![\"How](\"https://media.x.ai/cdn-cgi/image/fit=scale-down,onerror=redirect,width=6016,f=auto/v1/website/ksh-incidentbot-diagram-31b72f17.png\")
When a product incident occurs, the support queue can be one of the first places it shows up. Customers start reporting the same issue, sometimes before engineering alerts fire or anyone has declared an incident.
\nThose reports keep coming while the incident unfolds. At first, support may not know what is wrong or have an answer we can confidently provide. By the time the issue is resolved, hundreds or thousands of customers may be waiting for a response. Support has to track those customers through the incident, then make sure each gets the right answer when it is resolved.
\nWe use Grok Bot to handle both parts of this work. It provides us with AI teammates that take ownership of operational jobs, particularly repetitive workflows that otherwise fall between people and systems. As we shared in a recent blog, the SpaceXSI customer support team uses Support Bot in Grok Bot to handle customer-facing communication.
\nNow we’ve added two more AI teammates that we call on specifically for incidents. The first, IncidentWatch, monitors the support queue for spikes in similar tickets and alerts our team when it sees one. The second, IncidentSweep, handles the other end, identifying affected tickets and closing them in an approved batch.
\nDuring one incident on September 18, our Bot, IncidentSweep, cleared 204 tickets in about 10 minutes with a support engineer reviewing and approving the batch. Previously, we would have had to reply to those tickets one by one over the course of hours or use a bulk macro that gave us much less control over who received the message.
\n
Customers often notice problems before our internal alerts fire. When several customers report the same issue at once, those tickets can be an early signal of an incident.
\nIncidentWatch is a Grok Bot teammate we built to detect that signal. Every five minutes, it reviews the previous 15 minutes of tickets for spikes around the same problem. If the pattern does not match a known incident, IncidentWatch alerts the support engineer on call with the apparent issue, number of matching tickets, and examples to investigate.
\n![\"IncidentWatch](\"https://media.x.ai/cdn-cgi/image/fit=scale-down,onerror=redirect,width=6016,f=auto/v1/website/ksh-incidentwatch-9f4b8172.png\")
A support engineer decides whether the reports point to an incident and works with engineering to investigate. In the meantime, IncidentWatch holds tickets that clearly match the issue.
\nOnce the team confirms the incident, we post it to our public status page. Support Bot, our primary customer-facing Bot, returns to the held tickets, checks them against the incident, and responds to customers whose reports match. As the incident progresses, it can return with updates.
\nOnce an incident is resolved, there may still be hundreds of related tickets in the queue, and each needs a final response. Traditionally, support teams have had two ways to clear them, and both have drawbacks. They can respond to each ticket individually, which can take hours, or use a bulk action, which is fast but leaves little room to catch a ticket that should not receive the message.
\nWe built IncidentSweep to combine the control of the first approach with the speed of the second.
\nA support engineer starts by telling IncidentSweep which tickets to consider, which to exclude, and exactly what message to send. IncidentSweep finds the matching tickets, reviews each one, and sets aside any it cannot confidently connect to the incident.
\nBefore sending anything, IncidentSweep shows the engineer the proposed batch, the tickets it included, and the exact message they will receive. The engineer can spot-check the tickets and make changes before approving the send.
\nIncidentSweep then sends the approved message in small batches and records each action. It never adds tickets after approval, and if anything fails, it stops and reports what happened.
\n![\"IncidentSweep](\"https://media.x.ai/cdn-cgi/image/fit=scale-down,onerror=redirect,width=6016,f=auto/v1/website/ksh-incidentsweep-703c883b.png\")
Grok Bot introduces a different way of working with AI. Instead of turning to an AI tool for help with individual tasks, teams can create AI teammates that own jobs and carry entire workflows through to an outcome.
\nCustomer support has been a natural place for us to put this model to work. IncidentWatch and IncidentSweep take on jobs that used to require someone on the team to watch, sort, and act on hundreds of individual tickets. Across SpaceXSI, we’re looking for more workflows where AI teammates can take on the same kind of ownership.
\n", "date_published": "2026-10-20T00:00:00+00:00", "external_url": "https://x.ai/news/incident-sweep", "id": "urn:aggr:item:53bbfa766e9c9432b193a82ac7954f7d87d431ce", "image": "https://media.x.ai/cdn-cgi/image/fit=scale-down,onerror=redirect,width=6016,f=auto/v1/website/ksh-incident-og-0069bc57.png", "summary": "When a product incident occurs, the support queue can be one of the first places it shows up. Customers start reporting the same issue, sometimes before engineering alerts fire or anyone has declared an incident. Those reports keep coming…", "tags": [ "ai" ], "title": "Using Grok Bot to detect and respond to incidents", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-10-20-using-grok-bot-to-detect-and-respond-to-incidents/" }, { "content_html": "Today, we’re announcing Grok Voice 3.0, the world’s most intelligent speech-to-speech model. It sets a new record for resolving customer support calls, has class-leading latency at under one second, and costs less than other frontier models. Grok Voice powers millions of conversations a day across Tesla and SpaceX, and is available today in the API at the same price as its predecessor.
\nIn Artificial Analysis’s independent testing, Grok Voice 3.0 ranks #1 on the Speech-to-Speech Index, with a score of 86.9. The Index combines speech reasoning, τ-Voice, and preference and task success from blind head-to-head calls. Grok Voice 3.0 also scores 97% on Full Duplex Bench, a measure of how naturally it handles pauses, turn-taking, and interruptions.
\nSpeech-to-speech benchmark comparison between Grok Voice 3.0 and competing models, per Artificial Analysis.
\n| Benchmark | \nGrok Voice 3.0 (this release) | \nGrok Voice 2.0 | \nGPT-Live-1 (Astra, Medium) | \nGemini 3.8 Live (Ext. Thinking, High) | \n
|---|---|---|---|---|
| 86.9 | \n81.3 | \n83.2 | \n82.6 | \n\n |
| 98.5% | \n97.2% | \n90.1% | \n97.7% | \n\n |
| 1070 | \n1011 | \n1048 | \n990 | \n\n |
| 94.8% | \n94.6% | \n87.4% | \n89.1% | \n\n |
| 97.0% | \n95.1% | \n94.9% | \n91.9% | \n\n |
τ-Voice measures how well agents handle support calls with diverse accents, background noise, and interruptions. Grok Voice 3.0 scores 71.6%, comparable with GPT-Live-1 (Astra), but at 40% lower latency, setting a new frontier in agent intelligence versus responsiveness.
\nIt's also cheaper to run than comparable models. Artificial Analysis measures what each model costs to complete a task in its Big Bench Audio run, and Grok Voice 3.0 comes in under both GPT-Live-1 (Astra, Medium) and GPT-Realtime-2.1.
\nLower cost per task than comparable models
\nAverage cost per task on Artificial Analysis's Big Bench Audio run, USD. Lower is better.
\nOn EVA-Bench, ServiceNow’s benchmark for enterprise voice agents, Grok Voice 3.0 achieves the highest accuracy score at 74% and is the best-performing model when jointly scored across accuracy and customer experience.
\n\n\n“Grok Voice 3.0 sets a new Pareto frontier on EVA-Bench, ServiceNow’s open-source benchmark scoring voice agents on accuracy and experience across 200+ realistic enterprise scenarios. It pairs top accuracy with a natural, responsive conversation, setting a new standard for what enterprises can expect from AI on the front line of customer service.”
\n
Nitin Aggarwal, Senior Director of Product Management, Agentic Evals and RAI, ServiceNow
\nGrok Voice 3.0 handles millions of conversations a day in Tesla vehicles. Drivers can use Connectors for simple requests like checking their calendar or answering an email, and Grok Bot for bigger tasks like ordering a coffee or booking dinner—all completely hands-free (see it on X). Grok Voice 3.0 also powers Grok Bot’s voice mode.
\nStarlink call centers use Grok Voice to answer 20,000 customer calls every day across multiple languages. Roughly 70% are resolved with no human in the loop, and one in four sales calls ends in a purchase before the caller hangs up. A single agent built on Grok Voice 3.0 verifies customers, troubleshoots hardware, issues credits, and ships replacement equipment.
\nGrok Voice 3.0 is available today in the Voice Agent API at $0.08 / min ($4.80 / hr), the same price as Grok Voice 2.0. Agents using grok-voice-latest have already been switched over automatically, with no changes needed.
Today we’re introducing The X Subscription, which brings together Grok, Grok Bot, Cursor, and X Premium in one plan. Over the next few days, we’ll make it available to everyone globally.
\nFor the first time, one AI subscription supports the full path from thinking and creating to building and sharing.
\nPaid plans on The X Subscription include a generous pool of Grok model usage shared across Grok and Cursor. Each comparable new tier provides more usage-per-dollar than the plan it replaces.
\nYour shared pool covers Grok models whether you use them in the Grok apps, the Grok tab on X, or Cursor. Each X Subscription plan includes at least twice as much coding usage with our models as the comparable Cursor plan. With the new plans, third-party models are still available in Cursor on Premium and above. Usage is billed separately at API list prices, through top-ups.
\nThe X Subscription is available to new customers starting today.
\nWe’re also rolling out the option to switch for existing SuperGrok, Cursor, and X Premium subscribers. You’ll see it in your account billing settings when it’s available to you. You can keep your current plan, or switch and get a prorated refund for your old plan.
\nThe X Subscription gives you Grok, Cursor, and X for roughly what a comparable standalone plan for one of those products used to cost.
\n| Plan | \nPrice | \nWhat’s Included | \nIncluded Usage | \n
|---|---|---|---|
| Free | \n$0 | \nDaily access to Grok, a starter allowance for Cursor's coding agents, and standard X | \nLimited | \n
| Plus | \n$8/month | \nMore AI usage in Grok and Cursor, plus a checkmark, creator tools, and fewer ads on X | \nBase | \n
| Premium | \n$30/month | \nThe benefits of SuperGrok, Cursor Pro, and X Premium+, previously $90/month combined | \n4x Plus Plan | \n
| Super | \n$100/month | \nAll of the above features | \n15x Plus Plan | \n
| Ultra | \n$200/month | \nAll of the above features | \n50x Plus Plan | \n
For further details on pricing, included usage and eligibility see pricing page.
\nTeam users get Premium, Super, and Ultra on The X Subscription at the same per-seat prices as individual plans. That's priced lower per seat than older Cursor Teams but now includes 3x more coding usage, plus SSO, admin controls, consolidated billing, usage analytics, and SOC 2 compliance.
\nEnterprise customers can manage Grok and Cursor from one dashboard, with a single usage pool across both products. Our sales team is ready to help you get started.
\nNothing changes for existing SuperGrok, Cursor, or X Premium subscribers. You can keep your current plan, price, and features or move to The X Subscription.
\nSwitching comes with limited-time offers, which you can see in your account billing settings.
\nCompare plans and sign up at x.ai/pricing.
\nIf you already subscribe to SuperGrok, Cursor, or X Premium, go to your billing settings to see if switching is available for your account yet.
\n", "date_published": "2026-10-12T00:00:00+00:00", "external_url": "https://x.ai/news/x-subscription", "id": "urn:aggr:item:4c6af43562a138f8795670d1f7ef906c5f97b1c3", "image": "https://media.x.ai/cdn-cgi/image/fit=scale-down,onerror=redirect,width=6016,f=auto/v1/website/the-x-subscription-db7a1253.png", "summary": "Today we’re introducing The X Subscription, which brings together Grok, Grok Bot, Cursor, and X Premium in one plan. Over the next few days, we’ll make it available to everyone globally. For the first time, one AI subscription supports the…", "tags": [ "ai" ], "title": "Introducing The X Subscription", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-10-12-introducing-the-x-subscription/" }, { "content_html": "Today we’re launching Team Bots, Grok Bots that work and learn alongside your team. Give one access to the files, apps, and expertise it needs, then share it so everyone can work from the same context.
\nAt SpaceXAI, Team Bots brief account teams each morning, coordinate engineering projects, and answer data questions across the company. Here’s how they work, how we use them, and how to build one for your own workflow.
\nYou build a Team Bot around a role or workflow your team shares. Everyone gets access to the same Bot and its expertise, while still being able to work with it individually.
\nEvery Team Bot brings together four things that help it do its job:
\nAlthough the Bot is shared, each person’s conversations with it remain private. The Bot keeps separate context and memories for each user while drawing on the skills shared across the team.
\nTeams can also collaborate with a Team Bot in Slack. Each Team Bot has its own handle, so you can invite it to a channel where everyone can ask questions, contribute context, and see its responses.
\nAt SpaceXAI, we’re using Team Bots in some of our most context-heavy workflows. Here are four examples from Sales and Customer Success, Product and Engineering, Marketing, and Data Analytics.
\nAt SpaceXAI, every major sales account has a dedicated Team Bot. Each Bot is shared by the account executive, customer success manager, solutions architect, and sales leader.
\nEvery night, the Bot reviews company news, recent Gong calls, Notion docs, and relevant Slack threads. Each morning, it posts a briefing in the account’s Slack channel with what changed and what each person should do next, including drafts tailored to their role.
\nThroughout the day, account teams use the Bot in Slack to assess strategy, check its thinking against account data, and plan next steps. The Bot remembers the decisions teams make and builds rich context over time. As people rotate on and off the account, it becomes the system of record and can quickly bring new team members up to speed.
\nHarper, an insurance company serving small businesses, built a Team Bot to identify customers with lapsed policies and help them reinstate their coverage, reducing manual work for its team and recovering substantial savings for customers.
\n\n\n“We initially had to manually check every customer through three platforms, get their balance details, and send them personalized emails. We built a Team Bot in 24 hours to automate this, saving our customers over $120,000 from hundreds of policies. This is now an ongoing engine for us, saving us time and our customers' money.”
\n
Dakotah Rice, CEO, Harper
\nOur Engineering Team Bot works from a project’s Slack channel and connects to Notion, Linear, Hex, Datadog, and Cursor. It follows product decisions, triages bug reports, creates tickets, and launches Cloud Agents to handle well-defined fixes.
\nWe taught the Bot our shipping process through skills. It knows how to handle PR reviews, when to ask the team for help, and what evidence a change needs before it is complete. It coordinates the work across tools and agents, then reports progress and blockers in Slack.
\nWe used this setup while building Team Bots. The Bot steered a Cursor Project that orchestrated hundreds of Cloud Agents, helping a five-person team ship more than 100 PRs a day and launch Team Bots in a few weeks.
\nLarge marketing teams spend considerable time keeping their brand, messaging, and voice consistent. We built Marketing Bot to make that work easier.
\nWe gave the Bot access to our brand guidelines, blog posts, and social copy. Whenever someone shares a draft for approval in Slack, it reviews the work against our voice and latest messaging. Regional teams can get feedback and keep moving in their own time zones without waiting for an approval cycle at HQ.
\nFor website and SEO changes, Marketing Bot carries the work from review to launch. Once a content update passes review, the Bot makes the change and posts a preview link in Slack for final sign-off. Anyone on the marketing team can ship website updates without filing a ticket or waiting on engineering, and every change goes through the same brand and SEO checks.
\n\n\n“We’re building toward Team Bots for every marketing function. These Bots help us dig into market and customer intel, show up in AI answers, review launches, and tell us what’s working.”
\n
Angela Ferrante, Head of Marketing, Amplitude
\nOur data analytics team receives dozens of requests each day for one-off analyses. It built Data Bot so anyone at SpaceXAI can get answers without waiting for an analyst or setting up warehouse access.
\nData Bot uses shared, read-only credentials to query approved tables in Databricks. It also connects to Datadog, Hex, and Statsig to investigate questions and present the results.
\nThe data analytics team had already spent two years building a library of skills for working with more than 45,000 tables. It gave that existing library to Data Bot, including instructions for finding the right data, analyzing feature usage for fraud, following SpaceXAI’s chart standards, and safely updating company-wide dashboards. Data Bot also remembers corrections to its queries, so what it learns from one person improves the answers it gives everyone.
\nTeam Bots is available today in public beta on Teams and Enterprise plans. Take your favorite Grok Bots and share them with your team. To get started, try one of our pre-built Team Bots for sales, product management, marketing, or data analytics.
\n", "date_published": "2026-09-28T00:00:00+00:00", "external_url": "https://x.ai/news/team-bots", "id": "urn:aggr:item:7bd47a87cc91e4d0194b5452ae19483eb98f163c", "summary": "Today we’re launching Team Bots, Grok Bots that work and learn alongside your team. Give one access to the files, apps, and expertise it needs, then share it so everyone can work from the same context. At SpaceXAI, Team Bots brief account…", "tags": [ "ai" ], "title": "Team Bots: AI coworkers that learn from your team", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-09-28-team-bots-ai-coworkers-that-learn-from-your-team/" }, { "content_html": "When Cursor became part of SpaceXSI on August 14, our two customer support teams began coming together around a much broader product portfolio.
\nAt the same time, we were preparing to launch Grok Bot, an AI teammate you can give real work to. We expected the product to grow quickly, bringing another wave of users and support demand.
\nWe decided to use Grok Bot itself to help meet that demand, putting it to work throughout the support operation. It signed into the same tools our team used and its role stretched from resolving individual tickets to helping us understand and improve the operation as a whole.
\nOur new combined team has seen a 175% increase in support tickets, but we have not had to hire any new people thanks to Grok Bot. We might have hired 200 additional people otherwise.
\nWe are also doing it at a fraction of the usual cost. Traditional AI support tools charge a flat $1 to $4 per resolution. With Grok Bot, you only pay for your actual usage, which is already included in your plan. With minor optimizations, we've been able to resolve tickets for as low as $0.20 to $0.30.
\nWe took a crawl, walk, run approach to setting up Grok Bot. We started by connecting it to a few core systems, including Plain for ticketing and Linear for issue tracking. We then had it act as though it owned tickets, while limiting it to internal notes and requiring human approval for every write action. This let us check whether it understood each issue and proposed the right next step without affecting the customer experience.
\nAs the results became more reliable, we added traces and evaluations to every run. When something went wrong, we could see where Grok Bot had gone off course, make an adjustment, and try again. Grok Bot could also analyze these runs itself. This feedback loop allowed us to move quickly while keeping the process controlled.
\nOnce that foundation was in place, we began rolling Grok Bot out on the least complex tickets. During the first day, we manually reviewed its interpretation and proposed response for accuracy, tone, and whether it had followed our instructions. By the end of the day, we had enough confidence to let it begin responding directly to customers. From there, we gradually expanded the range of tickets it could handle.
\nIf you consider the end-to-end time that it takes to resolve a ticket, the majority of the clock happens during discovery, investigation, and troubleshooting. We began applying Grok Bot to every ticket as a pre-investigation step the moment it entered into our system. This could get expensive, so we've looked at common tickets and classified common issues to reduce the amount of tokens we needed to spend. We also don't exhaust a significant amount of troubleshooting capacity when a simple help center check does the trick.
\nWhenever we run into a known issue (it connects to our Linear instance), or if we hit a common error in our backend (it's connected to Datadog), we've trained Grok Bot to either add to the existing issue or to create a new one. Grok Bot also reproduces the issue with a video, which helps the engineering team quickly resolve it.
\nOf course, we also need to ensure that our customers are getting a clear response from us. We've trained Grok Bot on over one million customer interactions so that it's learned our tone and voice directly from our humans. Grok Bot is trained to not only respond, but always push the ticket towards resolution. It does this by asking relevant questions (i.e., it won't ask a question where the answer is already found in our logs).
\nGrok Bot can also take action on behalf of our customers. For example, we've provided it with clear refund instructions where 99% of all refund requests are resolved without human intervention.
\n![\"Grok](\"https://media.x.ai/cdn-cgi/image/fit=scale-down,onerror=redirect,width=6016,f=auto/v1/website/grok-bot-customer-support-refund-c952d80e.webp\")
Resolving individual tickets is only part of the job. We also need to understand what is happening across the queue. Grok Bot watches inbound volume continuously and adjusts the queue based on what needs attention. It can reprioritize tickets, reassign ownership based on urgency, and alert the organization when we are getting close to breaching a response-time SLA.
\nGrok Bot also looks across tickets for patterns. When the volume around a particular issue reaches a set threshold, it can declare an incident automatically. It monitors X for changes in sentiment and recurring reports of the same problem, giving us a view beyond the customers who contact support directly. Together, these signals help us spot emerging problems early.
\n![\"A](\"https://media.x.ai/cdn-cgi/image/fit=scale-down,onerror=redirect,width=6016,f=auto/v1/website/grok-bot-customer-support-incident-b3f2d249.webp\")
At our current scale, raw volume alerts would create a lot of noise. Grok Bot assesses whether a spike reflects a real support issue and begins investigating before it alerts the team. That gives us more context about what requires action while preserving the team's time and capacity.
\nAs Grok Bot took on more of our customer support work, it also gave us a new way to improve the operation itself. It reviews customer interactions handled by both people and Bots, provides specific feedback on what could be improved, and surfaces coaching opportunities for individual team members and Bots.
\nEvery week, Grok Bot sends our leadership team a summary of where our AI responses are falling short. Sometimes the answer is more training or better documentation. Other times, the summary confirms that the guardrails we put in place are working. This gives us a regularly updated view of quality and helps us address patterns early.
\nAs more users ask Grok for support, our help center increasingly serves as source material for its answers. To keep those answers accurate, Grok Bot reviews changes to our codebase and suggests corresponding updates to the help center.
\nWe have now reached the point where Grok Bots can coach other Grok Bots. They identify gaps in the knowledge system, fill those gaps, and feed what they learn back into the system. We are scaling this loop across the portfolio so that it covers every product surface.
\nGrok Bot has become our default data analyst, turning what it sees across customer support into daily reports for our Slack channels. When it detects early signs of a poor customer experience, it flags the situation so the team can step in while there is still time to change the outcome.
\nAn example of this is whenever a ticket goes back and forth more than three times between a customer and one of our team members (human or Bot). When that happens, Grok Bot flags the interaction for management review and gives leadership an opportunity to lean in and save the experience. This interaction helps train Grok Bot, and as Grok Bot learns which signals are useful to the team, the reporting becomes more relevant over time.
\nThe same analysis helps us improve product quality. Grok Bot synthesizes more than 20,000 points of product feedback from customer support tickets each day and turns them into clear themes we can bring to engineering. This gives the product team a broader view of where customers are struggling and where the product needs work.
\n![\"Grok](\"https://media.x.ai/cdn-cgi/image/fit=scale-down,onerror=redirect,width=6016,f=auto/v1/website/grok-bot-customer-support-themes-eb5918b6.webp\")
Grok Bot is still a new way of working, but it has already changed how our team operates. Instead of spending most of the day on repetitive support work, we can focus on setting guardrails, handling cases that require judgment, and deciding how the operation should improve. That makes the work more engaging and gives people more room to apply their experience to harder problems.
\nWe are still learning what this model makes possible. As the role of customer support continues to change, we will keep sharing what we find.
\n", "date_published": "2026-09-22T00:00:00+00:00", "external_url": "https://x.ai/news/grok-bot-customer-support", "id": "urn:aggr:item:d7eda8dce5adb990896e2644c1e65b97fd1df8b4", "image": "https://x.ai/images/news/grok-bot-customer-support-og.webp", "summary": "When Cursor became part of SpaceXSI on August 14, our two customer support teams began coming together around a much broader product portfolio. At the same time, we were preparing to launch Grok Bot, an AI teammate you can give real work…", "tags": [ "ai" ], "title": "How SpaceXSI is using Grok Bot to scale customer support", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-09-22-how-spacexsi-is-using-grok-bot-to-scale-customer-support/" }, { "content_html": "Grok 4.7 is our most capable model for coding and knowledge work. It works longer on difficult tasks, checks its own work more carefully, and comes with our best-calibrated safeguards to date. Served at the same price and speed as Grok 4.6, it is highly competitive in its class.
\nOn CursorBench 4.0, which stresses longer-running coding tasks, Grok 4.7 is at the frontier in price-performance.
\nGrok 4.7 uses a new, larger base model compared to Grok 4.6. It was trained with a longer reinforcement learning run on a harder mix of tasks, weighted toward problems that take many hours to complete. The model is better at verifying its own work and managing longer context. We also trained Grok 4.7 to natively understand the Grok Bot harness, making it better at conversational tasks and general knowledge work.
\nGrok 4.7 xHigh
\nGrok 4.6 High
\nGPT-5.6 Sol Max
\nFable 5.1 Max
\nInput token price $ per million
\n$2
\n$2
\n$4
\n$10
\nOutput token price $ per million
\n$6
\n$6
\n$20
\n$50
\nSoftware engineering CursorBench 4.0
\n46.3%
\n40.4%
\n41.7%
\n51.8%
\nSoftware engineering DeepSWE v1.1
\n71.0%*
\n65.2%
\n72.7%
\n70.0%
\nElectrical engineering EEBench
\n64.0%
\n53.0%
\n39.4%
\n56.4%
\nMulti-hour office work AA Briefcase v1.1
\n1,657
\n1,546
\n1,487
\n1,678
\nMulti-hour terminal work Terminal-Bench 4.0
\n37.6%
\n20.3%
\n37.3%
\n57.9%
\nLegal work Harvey Legal Agent Benchmark
\n19.6%
\n15.8%
\n2.5%
\n6.7%
\nClinical reasoning HealthBench Professional
\n56.7%
\n48.5%
\n60.5%
\n62.1%
\n* high effort
\nToken prices and benchmark scores for Grok 4.7, Grok 4.6, GPT-5.6 Sol, and Fable 5.1. Benchmarks are CursorBench 4.0, DeepSWE v1.1, EEBench, AA Briefcase v1.1, Terminal-Bench 4.0, Harvey Legal Agent Benchmark, and HealthBench Professional. An asterisk on Grok 4.7 DeepSWE marks a high-effort score.
\nGrok 4.7 is better at creating documents and presentations. In GDPval and AA Briefcase, AI is asked to work on tasks done by professionals such as lawyers, nurses, and financial analysts. Grok 4.7 improves upon Grok 4.6 on both benchmarks and performs comparably to other frontier models.
\nProfessional knowledge work
\nGDPval
\nGDPval, AA Briefcase, and EEBench scores comparing Grok 4.7 with Grok 4.6, Fable 5.1, and GPT-6 Astra.
\nGrok 4.7 was built with an entirely new safeguard stack. It is the strongest model we’ve tested on refusals and jailbreak resistance. In dual-use domains like cybersecurity and biological work, it leads on both utility for benign tasks and safe refusal on dangerous ones, topping LatchBio’s biosafety benchmark at 62.4%.
\nGrok 4.7 balances strong cyber defense capabilities with low refusal rates for legitimate use. It shows the highest safety on HackerBench v0.3, our benchmark for risky and malicious cyber tasks, allowing only 3.3% of risky dual-use prompts through while rarely blocking legitimate security work. We’ve also started giving select cybersecurity partners invite-only access to Grok 4.7’s red-team capabilities for defense research.
\nGrok 4.7 is available today in Cursor and Grok Build. It is also available through the Grok API, third-party coding harnesses, and model routers and cloud platforms.
\nThe model is priced starting at $2 per million input tokens and $6 per million output tokens. We also serve a fast variant with twice the output speed at twice the price.
\nGet started today at x.ai/build.
\n", "date_published": "2026-09-21T00:00:00+00:00", "external_url": "https://x.ai/news/grok-4-7", "id": "urn:aggr:item:847735558e3494faa1c392e58ff0f379fd040349", "image": "https://x.ai/images/news/grok-4-7-og.webp", "summary": "Grok 4.7 is our most capable model for coding and knowledge work. It works longer on difficult tasks, checks its own work more carefully, and comes with our best-calibrated safeguards to date. Served at the same price and speed as Grok…", "tags": [ "ai" ], "title": "Introducing Grok 4.7", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-09-21-introducing-grok-4-7/" }, { "content_html": "Today we're releasing Grok Voice Transcribe 2.0, our latest speech-to-text model. Across our real-world evaluations, Grok Voice Transcribe 2.0 is one of the most accurate transcription models available today and twice as accurate as Grok Voice Transcribe 1.0, at the same price.
\nGrok Voice Transcribe 2.0 is built on the audio foundation model behind Grok Voice. Grok Voice already powers tens of thousands of customer-support calls a day, transcribes millions of hours of video narration, and runs voice agents in physical products, including the Grok assistant in Tesla vehicles. It is trained on a unique dataset of live, noisy, multilingual audio recorded across a diverse set of environments and refined with post-training.
\nThe result is one of the most accurate transcription models for speech in real-world settings.
\nMost transcription models do well on clean, single-speaker audio. Real-world audio is harder: flaky phone lines, competing voices, local accents, and phone numbers or email addresses read aloud. We built Grok Voice Transcribe 2.0 for the hardest audio across conditions and environments.
\nOn the public Artificial Analysis leaderboard, Grok Voice Transcribe 2.0 ranks first for accuracy among 32 streaming models.
\nIn addition to public benchmarks, we measure word error rate on four internal sets drawn from production traffic: telephony audio from customer-support calls, conversations with Grok, spoken credentials such as account codes and email addresses, and short multilingual voice commands. Grok Voice Transcribe 2.0 improves on Grok Voice Transcribe 1.0 across all four, and on telephony it leads every model we tested.
\nGrok Voice Transcribe 2.0 transcribes dozens of languages, detects the language automatically, and follows mid-recording switches in a single pass. Multilingual accuracy is its largest improvement over Grok Voice Transcribe 1.0.
\nShort phrases such as in-car commands give the model little context to identify the language. On our short-phrase set, word error rate drops from 20.6% to 6.8%.
\nGrok Voice Transcribe 2.0 supports advanced configuration and controls. Existing Speech-to-Text API integrations get the accuracy improvement with no code changes:
\nAtlassian Loom is widely used for recording and sharing screen recordings. Atlassian found Grok Voice Transcribe 2.0 more accurate than their existing solution for transcribing Loom videos. Accurate transcripts open up new AI workflows: record an action plan in Loom, pipe the transcript into Cursor, and it makes the code updates directly.
\n\n\n“We've always believed the best way to move work forward is to capture context once and let it flow everywhere. With Grok powering Loom's speech-to-text and Cursor turning that into code, we're closing the loop from context to code: record what you mean, and the work gets done. It's a glimpse of where AI-assisted development is headed.”
\n
Sanchan Saxena, SVP of Teamwork Collection, Atlassian
\nGrok Voice Transcribe 2.0 pricing is identical to Grok Voice Transcribe 1.0. Batch transcription remains $0.10 per hour of audio and streaming $0.20 per hour, with diarization, timestamps, and key terms included.
\nGrok Voice Transcribe 2.0 will soon be the default in the Speech-to-Text API, and Grok Voice Transcribe 1.0 will be deprecated in the coming weeks. To stay on it during the transition, pin grok-voice-transcribe-1.0.
Grok Build now has memory. As you work, it keeps notes on the conventions, decisions, and project facts that come up, and later sessions read those notes before touching related code. Grok Build gets better the more you use it.
\nAfter a turn completes, Grok reviews it in the background and records anything durable: conventions, decisions, and project facts. Capture runs on every completed turn and does not interrupt the session.
\nMemory holds the details most likely to matter in a later session: how the team writes and reviews code, decisions and the reasoning behind them, and durable facts about the project, from where a subsystem lives to which command runs the suite. Task state, tentative conclusions, secrets, and anything the repository or its docs already cover are left out.
\nNotes are kept per project, plus a global set for preferences that apply everywhere.
\nCapture runs after a turn completes and never blocks the session. Grok reviews the finished turn, writes anything worth keeping as a markdown note, and continues. Over time, /dream folds those notes into topic files, one per subject, so each project ends up with a small set of organized references.
When you return to a project, Grok reads the topics that cover the area it is about to work in. Instructions in the current conversation take precedence over anything in a note.
\n/memory opens a read-only browser of every memory file, grouped by scope, with a preview of the selected file. It is the fastest way to see what a session produced, and to find the file to edit when a note is wrong.
/memory opens the browser above./dream organizes recent notes into topic files, for example topics/testing.md. Dream also runs on its own periodically in the background.Memory is available in Grok Build now. It applies to new sessions: run /new or start a fresh grok, and notes begin after the first completed turn.
Enterprises have a lot of spend that is hard to track closely. That includes unused SaaS seats that accumulate as teams change, and renewals that come up before anyone has checked whether the contract still matches how the product is being used. The same problem shows up in recurring purchases, where it is often easier to reorder what was bought last time than to adjust quantities to current needs or shop around for a better price.
\nA lot of this work is worth doing, but hard to justify doing by hand. We wanted to see how much of it Grok Bot could take on.
\nGrok Bot makes agents much easier to set up and use. You tell or show a Bot what you want it to do, give it access to the right tools, and let it execute without designing the workflow step by step.
\nFor procurement, we created a Bot we affectionately named Haggle Bot. It reads our vendor spend alongside contracts and usage data, then uses market pricing and competitive quotes to find savings and prepare negotiations. So far, it has identified more than $100,000 in direct savings, worked through larger SaaS renewals, and applied the same approach to recurring purchases like office supplies.
\nWe think Haggle Bot points to a broader role for Bots inside a company. Give a Bot a clear job and access to the tools it needs, and it can keep taking on the work within that role without being told each task.
\nThe goal of procurement is to support the business and get the most out of every dollar spent on vendors. A lot of existing procurement tools help with this at the vendor intake and contract-management level.
\nBut much of the most important procurement work hinges on questions those systems do not address:
\nThese questions are common, but depending on the stage of the company, they may not be worth answering by hand. Fast-growing companies often leave money on the table to avoid spending time chasing them down. That is where a Bot can be useful.
\nWe started Haggle Bot by describing the job we wanted it to do. Its instruction was to learn our vendor spend and turn that knowledge into evidence-backed savings, with a person making the final decisions.
\nProcurement
\nWe then gave the Bot access to Slack, Notion, Drive, Gmail, Hex, and Ramp. Haggle Bot used those systems to build a working map of roughly 125 active vendors.
\nThat record gave Haggle Bot enough context to make decisions and keep working without a person spelling out each next step. We also defined where Haggle Bot should stop and hand back control. We let it handle internal research and coordination on its own, but required explicit approval for spending money, accepting terms, or sending something to a vendor.
\nAn easy place to start was auditing SaaS spend. Haggle Bot asked our IT team for assigned-seat and last-used data, then compared that with what we were paying for. For one product, it found 43 paid seats with no activity in the previous 90 days and sent the names back for review and downgrade. That added up to $14,220 in savings.
\nWe applied the same approach to another SaaS product and Haggle Bot found $85,662 a year in unused SKUs. Because the product was month-to-month, those cuts reduced spend immediately.
\nOne thing we noticed in these SaaS audits was how often Haggle Bot took the next step without being asked. In one case, Haggle Bot needed to figure out who owned the relationship and what our plans were for the vendor. It started with the owners listed in Ramp, messaged them, and followed each handoff until it reached the engineers with the right context to make a decision. Rather than stop when the first answer was incomplete, it identified what additional information it needed and then proactively sought it out.
\nWhen a SaaS renewal came up, we asked Haggle Bot to review the quote and explore alternatives. Haggle Bot compared the renewal offer with our current annualized spend and recommended against the options that added seats ahead of demonstrated usage.
\n![\"Haggle](\"https://media.x.ai/cdn-cgi/image/fit=scale-down,onerror=redirect,width=6016,f=auto/v1/website/haggle-bot-image1-75f1097a.png\")
It then priced credible alternatives against our current footprint and used those comparisons alongside current usage data to work out where we had negotiating leverage.
\n![\"Haggle](\"https://media.x.ai/cdn-cgi/image/fit=scale-down,onerror=redirect,width=6016,f=auto/v1/website/haggle-bot-image2-7f0e2a56.png\")
From there, Haggle Bot worked out the negotiation and drafted the response for us to edit. We set the minimum quantities we wanted to keep and approved the send. Haggle Bot made an opening bid while setting an internal price target we were willing to go up to.
\nEvery Friday, our office team orders tech, snacks, and hygiene supplies for the next group of new hires. We use another Bot to place that order, which we call "Amazon Bot." It's logged into our corporate account and can work across Gmail, Ramp, Google Sheets, Rippling, and Vercel to understand headcount and the office floor plan.
\n![\"Amazon](\"https://media.x.ai/cdn-cgi/image/fit=scale-down,onerror=redirect,width=6016,f=auto/v1/website/haggle-bot-image3-b5f9709c.png\")
Rather than treat that as a fixed Amazon order, we gave Haggle Bot the job of shopping the weekly order around. Haggle Bot can see how quickly supplies are being used, the seat map for each building, and the quotes and carts from the last four orders. From that, it builds an editable Google Sheet where office ops can change the number of incoming hires and see the quantities needed for each new-hire kit.
\nHaggle Bot then shops the order across Amazon, Costco, Uline, and Walmart. If it cannot find the same product for less, it can look for an equivalent from another brand, then collect the comparisons in a sheet for review.
\nIt then drafts an email to our Amazon procurement rep with same-day competitor prices and asks for lower pricing on specific line items through our business discount program.
\n![\"Haggle](\"https://media.x.ai/cdn-cgi/image/fit=scale-down,onerror=redirect,width=6016,f=auto/v1/website/haggle-bot-image6-f62bac1c.png\")
Once pricing is settled, Haggle Bot delegates the order back to Amazon Bot to send. In one run, the process brought a $14,629 tech order down to $6,143, a 58% reduction.
\nHaggle Bot shows how much work can be done behind a simple expression of intent. Once a Bot has a purpose and the access to pursue it, it can keep making progress across systems without additional intervention.
\nHaggle Bot has also helped us see where Bots fit naturally into a workflow, and where they need more guidance. A lot of procurement work depends on good judgment about how to communicate with vendors, and we still revise its emails to calibrate on tone and make sure we're providing vendors with the right level of information.
\nHaggle Bot has made us interested in what Bots can do when they keep working on the same job over time. The examples above came from a short window, but many of the signals worth acting on only emerge gradually as spend and usage change. Over the next year, we expect Haggle Bot to find more of that work on its own and take it further before a person needs to step in.
\nYou can copy and import Haggle Bot now in the Bot Marketplace.
\nJoin the Grok Bot for Enterprise waitlist.
\n", "date_published": "2026-09-04T00:00:00+00:00", "external_url": "https://x.ai/news/grok-bot-procurement", "id": "urn:aggr:item:10d384c3d8b6777e39de8b199aa1ee75346e120c", "image": "https://x.ai/images/news/grok-bot-procurement-og-split.webp", "summary": "We gave Grok Bot access to vendor spend, contracts, and usage data. It found more than $100,000 in direct savings.", "tags": [ "ai" ], "title": "Setting Grok Bot loose on procurement", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-09-04-setting-grok-bot-loose-on-procurement/" }, { "content_html": "Grok Bot is your team of helpful AI teammates. You delegate real tasks to them and they carry the job through end to end, working autonomously around the clock inside the same tools you use.
\nEnterprises need the ability to govern Bots at scale and today's release adds access, network, and audit controls that make that possible.
\nA Bot is a worker you create inside Grok Bot for a specific job. Each Bot runs on its own computer in the cloud and can use every app and website the same way you do. You message a Bot like a coworker, and it comes back when the work is done or it needs a decision from you.
\nYou'll usually manage several Bots for different jobs and each runs independently. To teach a Bot a workflow, have it follow along once. It saves the routine, takes your corrections, and runs it on its own from then on. When you've built a Bot that works well, you can hand it to the person next to you as a template. Bots can also message each other and share context, so you're not having to pass context between them.
\nThousands of organizations have adopted Grok Bot since launch, including customers like Legora, Supermicro, and ServiceTitan. These teams use Grok Bot every day, and the heaviest use is outside engineering. Here are a few popular examples of the [millions] of bots created in the past few weeks:
\nSales. Bots can watch webinars and podcasts by people at target accounts, then leave LinkedIn and email drafts for morning review. During a call, a Bot updates the deck from live notes so next steps are on the slide even before the meeting is over.
\nRecruiting. A Bot prospects overnight and builds a morning shortlist with outreach queued for review. It texts candidates good luck before they walk in, and after interviews it submits scorecards from that day's Gong calls.
\nMarketing. After a webinar, a Bot pulls the Zoom Q&A and sends Slack messages to the AEs whose customers showed up. The messages include notes on the questions the customer asked as well as draft replies.
\nFinance. A specialized procurement Bot monitors vendor spend, usage, and renewals, then finds places to cut costs or negotiate better terms. It has surfaced tens of thousands of dollars in savings across SaaS, recurring purchases, and vendor services.
\nEngineering. Bots monitor PRs for bugs, security findings, failing builds, and merge conflicts, then keep every task moving until it's ready for review. Overnight audits improve code quality, performance, and security, giving engineers more time for the hardest problems.
\nEach user's work in Grok Bot runs in its own secure and isolated environment, separate from every other user. A Bot has no access by default and reaches only the accounts you sign it into.
\n\nGrok and Cursor Enterprise customers get Grok Bot free for the next two weeks and can invite their whole organization, including people without an existing seat.
\nActivate Grok Bot today in your admin dashboard.
\nTo learn more, contact sales.
\n", "date_published": "2026-09-03T00:00:00+00:00", "external_url": "https://x.ai/news/grok-bot-for-enterprise", "id": "urn:aggr:item:88b8b6c4d35bf75092cddb6675953c227ca97248", "image": "https://x.ai/images/news/grok-bot-for-enterprise-og-split.webp", "summary": "Grok Bot is now available for enterprises. Grok and Cursor Enterprise customers have free usage for the next two weeks, and can invite their whole organization, including people without an existing seat.", "tags": [ "ai" ], "title": "Grok Bot for Enterprise", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-09-03-grok-bot-for-enterprise/" }, { "content_html": "When we started designing Grok Bot, one of the central questions was how the interface should shape the relationship between user and agent. Most AI interfaces are organized around a chat session the user operates. Each session begins with setup, unfolds as the user looks on, and ends when the conversation stops.
\nWe wanted to design for an agent that persists beyond any one session and can carry responsibility on its own. That meant reconsidering some of the basic objects and signals of the interface, including what belongs in the sidebar, how an agent shows progress, and when its work should become visible.
\nAI products have accumulated a large vocabulary in a short time. Chats, sessions, models, context windows, memories, system prompts, projects, skills, connectors, agents, tools, sandboxes, permissions, and automations all describe real parts of these systems.
\nBut exposing each one as a separate product concept asks users to understand more than they need to. We started by asking which concepts a person actually needs in order to work with an agent.
\nWe kept coming back to five:
\nEverything else could remain beneath the interface until the user had a reason to care about it. The next question was which of these five objects should organize the product.
\nChats are disposable. We start a conversation to solve a problem. It gets pushed down the sidebar. A week later, we start another one. You rarely go back beyond the most recent five.
\nThat behavior is perfectly reasonable when the unit of interaction is a question. It becomes strange when the thing on the other side of the interaction is supposed to know you, remember previous work, and take responsibility over time.
\nSo the main objects in Grok Bot are Bots, not conversations. A Bot has a name. It has an avatar and a title. It remembers its conversations with you. It has its own computer and tools. When you come back tomorrow, you are coming back to the same Bot.
\nOnce a Bot was something you maintain over time rather than a session you start, the way Bots appear in the product had to answer three questions at once:
\nA roster only works if it can be scanned quickly. As the roster grows, we did not want people to have to read every name each time they opened the product. They should be able to recognize a Bot from its avatar almost peripherally.
\nBot avatar visual style explorations
\nAt the same time, we wanted to keep the avatars consistent enough to read as one system. We studied character systems across illustration, animation, games, and interface design, exploring everything from initials and emojis to pixel art, watercolor, claymorphism, Noritake-style line art, silhouettes, and identicons.
\nMost approaches solved one side of the problem better than the other. Watercolor and clay gave individual Bots plenty of character but carried too much detail at sidebar scale. Simpler systems sat more naturally in the interface, but often left the Bots looking interchangeable.
\nThe system we landed on keeps the basic construction consistent, using simple shapes and expressive eyes, then introduces distinction through controlled variations and accessories. Each Bot remains recognizable at a glance without appearing to come from a different visual world.
\nOnce the avatar became the Bot’s identity, it was also the natural place to show state. A Bot may be idle, thinking, working, waiting, blocked, or done. We could have represented each state with a separate indicator, but that would have added another layer of UI for the user to interpret.
\nInstead, we explored how much of the lifecycle the avatar itself could carry.
\nAt rest, the Bot is calm and slightly curious. When work arrives, it acknowledges the task. As work begins, it kicks into gear. Its motion changes again when it is waiting or needs help, then settles once the work is done. The avatar now shows what the Bot is doing as well as which Bot it is.
\nAvatar motion system
\nA related design question was how much of the Bot’s execution to show. One approach would have been the standard “three animated dots” but that would have been too little information, making it hard for users to tell whether the Bot was working or stuck.
\nWe also tried showing a short written description of the Bot’s current action, but once people could see one step, they wanted to see the rest. User research showed us that they were asking for that detail mainly for reassurance that the Bot was still working and on the right track.
\nIn the final design, the avatar’s motion provides the first bit of reassurance by showing that the Bot is active. If someone wants to check what it is doing, they can hover to see its current action.
\nEach Bot has its own computer, which it can use to browse the web, work with files, and run software. This created another interface problem. How visible should that computer be and when should the user be able to control it?
\nWe explored four arrangements:
\nThe more prominent we made the computer, the more the product encouraged users to supervise it. We decided it should remain the Bot’s workspace, with the interface providing different levels of access as the user needed them.
\nThe final design has three levels, which allow the user to enter the Bot’s workspace without being drawn into operating it:
\nWe also designed wallpapers that shift throughout the day, becoming lighter in the morning and darker at night. The detail gives the Bot’s computer its own sense of time and makes it feel separate from the user’s desktop.
\nDynamic wallpaper
\nIt is closer to working with a coworker than operating a remote machine. You can tell that they are working, glance at their screen when you need context, and sit down when something requires your help.
\nEarly versions of Grok Bot responded to almost every request with prose. It described a five-day forecast instead of showing one and narrated a set of tasks instead of laying them out as a board. The user then had to restructure the answer. This led us to treat the form of a response as part of the answer.
\nTo support this, we built inline cards and widgets into Grok Bot. A Bot can answer in prose when prose fits the information and use structured UI when it does not.
\nInline chat widgets
\nThe same principle applies to actions. When a Bot creates a Routine, changes a setting, or messages another Bot, the event can appear directly in the transcript. The user can open it when there is more to inspect.
\nThe result is a heterogeneous transcript in which conversation, system events, interactive objects, and visualizations share one timeline.
\nOnce people create several Bots, the product also has to organize how those Bots work together. We needed to decide which context should belong to each role, how Bots should share context when their work overlaps, and how to coordinate them without turning the user into a dispatcher.
\nWe saw one answer emerge as people created more Bots. Some made a Chief of Staff Bot responsible for coordinating several specialists. They could give direction to one Bot instead of checking each one and routing every task themselves.
\nGiving Bots distinct roles also forced us to decide what each role should know. A legal Bot may need the history of an ongoing dispute, while a finance Bot may need years of financial records. Combining those histories into one large memory would make it harder to give each Bot the information relevant to its work.
\nCapabilities and context therefore follow different boundaries in Grok Bot. Tools and Skills live at the account level because many Bots may need to browse the web, work with documents, or send email. Memory and Routines belong to the Bot because they reflect what that particular role knows and does over time. Put another way, capabilities can be shared broadly while context remains with the role that needs it.
\nSome work crosses those role boundaries. Group chats provide shared context for a project or team while allowing each Bot to retain its specialized memory. A designer, engineer, PM, and data scientist can work in the same conversation, hand work to one another, and share what the project requires.
\nWe considered adding dashboards, assignment boards, and explicit handoff controls to manage these groups. Each one gave the user more coordination work. Instead, coordinating Bots handle routine routing and bring the user in when a decision requires judgment.
\nMost agent sessions begin when a user sends a prompt. That leaves even a persistent Bot waiting for someone to activate it. Routines let users give a Bot a standing responsibility that runs on a schedule or in response to an event, such as watching an industry or preparing a briefing every morning. The user defines the work once, and the Routine activates the Bot when it needs to happen.
\nWe initially treated Routines as secondary configuration. As they became more important to autonomous work, we moved them into the Bot’s main interface. The transcript shows what ran and gives the user a place to review the result or handle an exception.
\nThis also changes the role of conversation. A prompt can start a session, but so can a schedule, an event, or another Bot. Over time, more work may begin without the user being present at all.
\nBy the end of the project, much of the design work involved taking things away. We removed window and panel controls, computer-view options, and agent metadata. We also set practical limits of roughly 50 Bots per account and six per group chat. Each decision came back to the same question: Did this help someone delegate, or did it give them one more thing to manage?
\nThe line between operating an AI and delegating to a coworker keeps moving as models improve. Grok Bot reflects where we think it sits today. Designing Grok Bot from its earliest explorations through launch has been about finding that line and helping the interface change with it. As agents take on more responsibility, the interface should ask less of the person.
\n", "date_published": "2026-09-03T00:00:00+00:00", "external_url": "https://x.ai/news/designing-grok-bot", "id": "urn:aggr:item:28d5a4b714f9c48b3f2ca2beff96e12654a161ab", "image": "https://x.ai/images/news/designing-grok-bot-og.png", "summary": "How we designed Grok Bot for agents that persist beyond a single session — from a chat history to a Bot roster, presence, a computer of the Bot’s own, and work that starts without a prompt.", "tags": [ "ai" ], "title": "Designing Grok Bot for a world of persistent agents", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-09-03-designing-grok-bot-for-a-world-of-persistent-agents/" }, { "content_html": "Today, LatchBio published an independent analysis of Grok 4.6 on their biological capability and biological red-teaming benchmark suites. These evaluations measure a wide range of characteristics and capabilities of tested models, including:
\nOn LatchBio's BioSecBench-Refusal suite, Grok 4.6 was the strongest model tested at refusing disguised and hazardous tasks while still completing routine biological work. It was the only system to score above 50% on both measures.
\nBiology and biological research present domains where the opportunities and risks of highly capable AI systems are well-known and challenging to disentangle. In designing our safeguards, we aim to maximize the utility of our agents in assisting and accelerating legitimate scientific research, while minimizing the risks they present through incorrect information and adversarial use.
\nLatchBio's two benchmarks most relevant to the above capabilities measure whether an agent can distinguish concealed hazards from ordinary science, and whether it can carry out the pathogen surveillance workflows public-health work depends on. They are:
\nIn testing on BioSecBench-Refusal, LatchBio found that Grok 4.6 detects and refuses red-team and otherwise dangerous queries more reliably than any other frontier system tested, while not diminishing performance on routine biological work. Grok 4.6 also performs comparably with other frontier models on biosurveillance work. LatchBio notes that evaluations were performed on a variety of agent harnesses to remove confounds, and that unless otherwise noted, agents were tested at their highest-offered effort levels.
\nThe score presented by LatchBio is a trial-weighted harmonic mean of red-team refusal and routine compliance. We analyze this metric and standalone refusal rate independently. Across different harnesses, Grok 4.6 holds the top three spots, averaging 62.1%. Considering refusals and task compliance independently, Grok 4.6 refused 59.2% of red-team tasks and completed 64.8% of routine ones. It is the only model tested that scored above 50% on both measures.
\nOn BioSecBench-Surveillance, Grok 4.6 averages a success rate of 53.5%, sitting behind Opus 5 and ahead of GPT-5.6 Sol on biosecurity and monitoring work. The results on both benchmarks indicate an agent and underlying model well-calibrated for routine and helpful biological work, as well as one highly capable in biosecurity work and monitoring.
\nOn evaluations of general routine biological capability performed by LatchBio (such as SpatialBench or TxBench-PP), Grok is found to match or exceed other frontier models in a wide range of agentic biological work. These are presented in greater detail at benchmarks.bio.
\nIn evaluation traces, Grok 4.6 is observed reasoning over the contents of a task and testing environment to assess intent before proceeding or refusing. Frequently, Grok will find discrepancies between the stated intent in the prompt and environment, or will assemble intent from high-risk content disguised by filenames and encryption, and will subsequently refuse. On obviously benign and low-risk tasks, Grok exhibits the same environment-reasoning behavior, but is able to assess tasks as safe.
\nBefore we release a model, we test biological capability in addition to other domains of risk, and validate that the capabilities of our models are adequately safeguarded against misuse. The work of third-party evaluators such as LatchBio complements the evaluations we perform internally, both pre- and post-deployment, for the models we serve.
\nGrok's safeguards are built in layers to establish defense-in-depth. Refusal training is performed to teach the model how and when to refuse, to correctly infer the intent and risk profile of tasks, and to refuse correctly in highly adversarial scenarios. We train and deploy inference-time safeguards to reject harmful requests before they ever reach the model, and we implement behavioral controls to further safeguard the model when deployed. Post-deployment monitoring is performed to detect and stop patterns of adversarial use at the session and user level, and provides a source of feedback for continuous calibration of our model deployments.
\nWe observe a material improvement in Grok 4.6's capabilities in biological work over historically tested models, with substantial gains in refusal and biosecurity performance over Grok 4.5 and Grok 4.3. We intend our ongoing and novel safeguards work to allow us to continue serving frontier-scale intelligence safely across future model releases.
\nThe line between actively adversarial tasks and helpful use grows thinner as models rapidly become more capable and autonomous, and safeguards will need to evolve accordingly. To control for the greater risk implied by improved model capability and agency, we will perform wider, more ambitious testing and calibration of our models and agentic systems: broader pre-deployment suites, more third-party evaluations, improved post-deployment monitoring, and deployments of our models with companies and institutions at the frontier of biology.
\nSimilarly, there is a risk inherent in overrefusals and miscalibrated safeguards. When a model refuses routine and helpful biological work, the ability of healthcare professionals, researchers, and monitoring programs to detect outbreaks early and perform other critical work in the field is degraded. We gauge this risk as equally serious as the risk of aiding malicious use.
\nGrok is already used in scientific work, including general biological research and biosecurity monitoring, and we are optimistic about the ability of highly capable models to rapidly accelerate scientific discovery and shorten the path from the lab to practical use. We are committed to continuing to serve frontier intelligence securely and safely to the engineers, researchers, enterprises, and institutions advancing the life sciences.
\nLatchBio's full methods and per-model scores are presented at benchmarks.bio. LatchBio's complementary blog, the Grok 4.6 model card, and SpaceXSI's Frontier Artificial Intelligence Framework are linked below.
\n\n", "date_published": "2026-09-01T00:00:00+00:00", "external_url": "https://x.ai/news/biosafety-at-the-frontier", "id": "urn:aggr:item:198cd2f3f6559d7892f37876f187140b5d5611c6", "image": "https://media.x.ai/v1/website/biosec_frontier-1584260e.png", "summary": "LatchBio evaluated Grok's performance on biosecurity monitoring and adversarial biological tasks. They found that Grok 4.6 detects and refuses dangerous queries more reliably than any other frontier system.", "tags": [ "ai" ], "title": "Biosecurity at the frontier", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-09-01-biosecurity-at-the-frontier/" }, { "content_html": "Grok Bot now has a tighter integration with X.
\n", "date_published": "2026-08-29T00:00:00+00:00", "external_url": "https://x.ai/news/grok-bot-and-x", "id": "urn:aggr:item:2b2e13a4bb7070d3b585791ecfc9a6fec820b36b", "image": "https://x.ai/images/news/grok-bot-and-x-og.webp", "summary": "Grok Bot now has a tighter integration with X.", "tags": [ "ai" ], "title": "Grok Bot now works with X", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-08-29-grok-bot-now-works-with-x/" }, { "content_html": "We launched Grok Bot in beta on August 11 to give people highly capable AI teammates who could get real jobs done and be incredibly easy to work with.
\nSince then, people who tried Grok Bot were surprised at how easy it was to start, and how much sharper Bots felt than other AI tools, taking jobs further without being walked through every step.
\nToday, we're expanding access to Grok Bot. Grok Bot is now included with all SuperGrok, Cursor Pro, and Cursor Teams plans.
\nWith this expansion, Grok Bot is now included with:
\nGrok Bot comes with its own usage, separate from your Grok and Cursor plans, so anything you hand off to a Bot won't count against your existing usage.
\nGrok Bot is designed for people who want agents to handle anything from a project to an entire function. They're digital teammates you can message, run in parallel, and trust to finish work in the tools you already use. Bots work across apps and inboxes, keep going when you step away, and only pull you in for judgment calls.
\nWith Grok Bot, you get:
\nDownload Grok Bot so you can create your first teammate and give it a job you already do. You'll see what useful AI looks like when the work comes back finished.
\nEnterprise users can join a waitlist as we ramp access for larger team and company rollouts.
\nRead the original announcement and see what Bots can do to learn more.
\n", "date_published": "2026-08-26T00:00:00+00:00", "external_url": "https://x.ai/news/grok-bot-more-plans", "id": "urn:aggr:item:dd08dbb61faa3c84ed34751a6c9c1e29a34defbe", "image": "https://x.ai/images/news/grok-bot-more-plans-og.webp", "summary": "Grok Bot is now available for SuperGrok, Cursor Pro, and all Cursor Teams plans.", "tags": [ "ai" ], "title": "Grok Bot is now included with more plans", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-08-26-grok-bot-is-now-included-with-more-plans/" }, { "content_html": "Grok 4.6 is now available via Microsoft Foundry.
\n", "date_published": "2026-08-26T00:00:00+00:00", "external_url": "https://x.ai/news/grok-4-6-microsoft-foundry", "id": "urn:aggr:item:8196d7b6296f20e28c749d555eb541a4f683b8fb", "image": "https://x.ai/images/news/og-4-6-on-foundry.webp", "summary": "Grok 4.6 is now available via Microsoft Foundry.", "tags": [ "ai" ], "title": "Grok 4.6 on Microsoft Foundry", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-08-26-grok-4-6-on-microsoft-foundry/" }, { "content_html": "Grok 4.6 is now available via Gemini Enterprise Agent Platform.
\n", "date_published": "2026-08-21T00:00:00+00:00", "external_url": "https://x.ai/news/grok-4-6-vertex-ai", "id": "urn:aggr:item:458e1c3687e00890f6246af3de133fdfeea3516d", "image": "https://x.ai/images/news/og-gemini-enterprise-agent-platform.webp", "summary": "Grok 4.6 is now available via Gemini Enterprise Agent Platform.", "tags": [ "ai" ], "title": "Grok 4.6 on Gemini Enterprise Agent Platform", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-08-21-grok-4-6-on-gemini-enterprise-agent-platform/" }, { "content_html": "Describe an app, a game, a website, or a dashboard in Grok, and Grok builds a working version live in your chat — on the web, on iOS, and on Android.
\nWhen we introduced it in July, building in Grok was an Early Beta limited to SuperGrok Heavy. Since then we've made it faster, taught it to publish and share what you make, wired it into X, and given the apps it builds access to our own models. All of that ships to everyone today.
\nEvery example below is a real published app, live at its own grok.me link, created from the message shown above it.
\nBuild a peaceful 3D driving game: an endless forest I can drive through freely, with no missions and no timers. Make it feel calm.
\nEvery app you publish gets its own address on grok.me, and you decide who can open it: just you, anyone with the link, or the whole internet. Bring your own domain if you'd rather.
Grok also creates the art that travels with the link. Each app gets a cover that unfurls as a rich preview wherever it's pasted — messages, chats, docs, the timeline. It's generated for your specific app, and you can regenerate it, upload your own, or ask Grok to redesign it mid-conversation.
\nIllustration of the publish sheet. The cover is generated for you and can be regenerated or replaced.
\nApps you share on X don't arrive as bare links. The banner renders inline, carries your handle, and games get a play affordance in the card, so people meet the app before they ever leave the feed.
\nTurn on the SpaceXSI APIs for an app and it can call Grok for chat, images, and voice from inside its own code — no keys to create, paste, or rotate. A trivia game can write new questions on demand, a support page can answer in your voice, a notes app can summarize itself. Access is per app, and you can revoke it at any time.
\nAlso new since the Early Beta:
\nOpen Grok on the web, iOS, or Android, and describe what you want.
\nGrok 4.6 is now available via Amazon Bedrock.
\n", "date_published": "2026-08-19T00:00:00+00:00", "external_url": "https://x.ai/news/grok-4-6-amazon-bedrock", "id": "urn:aggr:item:788a0334dfd737e5e2f7601a1fc15fb60a0f74ed", "image": "https://x.ai/images/news/og-grok-4-6-on-bedrock.webp", "summary": "Grok 4.6 is now available via Amazon Bedrock.", "tags": [ "ai" ], "title": "Grok 4.6 on Amazon Bedrock", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-08-19-grok-4-6-on-amazon-bedrock/" }, { "content_html": "Grok 4.6, our latest coding model, is now available in GitHub Copilot.
\n", "date_published": "2026-08-14T00:00:00+00:00", "external_url": "https://x.ai/news/grok-4-6-github-copilot", "id": "urn:aggr:item:60ae716f8d97a48ac603f76fbce2237aa77b11ec", "image": "https://x.ai/images/news/og-4-6-in-github-copilot.webp", "summary": "Grok 4.6, our latest coding model, is now available in GitHub Copilot.", "tags": [ "ai" ], "title": "Grok 4.6 in GitHub Copilot", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-08-14-grok-4-6-in-github-copilot/" }, { "content_html": "Today we are releasing Grok 4.6. Grok 4.6 builds on Grok 4.5 with a particular focus on long-running agents and more ambitious interactive and visual work. It stays with complex tasks across many steps, whether researching a topic, analyzing information, working across a codebase, or turning an idea into a polished application or work artifact.
\nGrok 4.6 achieves frontier intelligence across several agentic coding and knowledge work benchmarks. It matches GPT-5.6 Sol on the Artificial Analysis Intelligence Index, which is a composite score of nine benchmarks.
\nCompetitor figures are drawn from the respective developers’ published system cards or benchmark leaderboards
\nBenchmark bar charts comparing Grok 4.6 with other leading models across AA Intelligence, GDPVal-AA, DeepSWE 1.1, CursorBench 3.2, and FrontierCode 1.1. Competitor figures are drawn from the respective developers' published system cards or benchmark leaderboards.
\nGrok 4.6 is available today in Cursor and Grok Build. We’re offering 2x included usage inside Grok Build and Cursor for the first week so you can start trying 4.6 immediately.
\nGrok 4.6 underwent a longer supplemental training run than Grok 4.5, with curated model-generated data for reasoning and advanced technical concepts, high-quality engineering data, and an improved optimizer and training recipe. This produced a stronger foundation for the SFT and RL stages that followed.
\nWe then used Grok 4.5 to regenerate the SFT trajectories across reasoning efforts, agent harnesses, and domains such as STEM, software engineering, and knowledge work, and filtered out problematic traces with model-based checks. The resulting SFT checkpoint shows strong performance and improved behavior.
\nGrok 4.6 is trained on a wide range of agentic RL tasks, including knowledge work, general coding, and domain-specific environments for kernel optimization, web development, computer-aided design, and more.
\nWe tested Grok 4.6 on projects designed to stretch its range and ability to sustain work over many steps. We found the model is especially strong at turning a broad product idea into a working first version. It can research unfamiliar domains, structure the application, implement the core interactions, and continue refining the result through several rounds of feedback.
\nOn longer trajectories, we also started to see more self-testing and verification, with the model checking its own work before moving on.
\nGrok 4.6 produces stronger first passes on visual and interactive projects than we typically saw with Grok 4.5. Given a concrete product idea, it is able to establish structure and visual language for an application in one pass. This has made it especially useful for projects where the fastest route to a good result was to begin with something substantial and then iterate in the loop.
\nGrok 4.6’s safeguards have been improved and calibrated in line with the model’s capabilities.
\nOur safety stack is designed to maximize utility and security across legitimate use cases, allowing Grok 4.6 to be helpful and safe in domains such as vulnerability patching, accelerating the engineering design cycle, and augmenting AI research.
\nOur safeguard evaluation work reflects Grok 4.6’s expanded capabilities, with our widest-ever suite of pre-deployment testing for capabilities and safeguard calibration, as well as extensive post-deployment and third-party testing.
\nGrok 4.6 High
\nGrok 4.5 High
\nGPT-5.6 Sol Max
\nFable 5 Max
\nAA Intelligence Index
\n61
\n56
\n61
\n62
\nGDPVal-AA v2
\n1753
\n1526
\n1728
\n1741
\nCursorBench v3.2
\n69.9%
\n66.7%
\n67.2%
\n70.5%
\nDeepSWE v1.1
\n65.9%
\n54%
\n73%
\n70%
\nFrontierCode v1.1 (Extended)
\n61.3%
\n56.6%
\n60.6%
\n63.6%
\nAPEX-Agents
\n57.5%
\n47.1%
\n56.7%
\n59.2%
\nTerminal-Bench v3.0
\n26%
\n15.7%
\n34.6%
\n34.1%
\nAPEX-SWE
\n56.4%
\n53.6%
\n—
\n58.8%
\nAA-Briefcase
\n1577
\n1313
\n1502
\n1574
\nHarvey LAB (Vals)
\n15.8%
\n12.9%
\n2.5%
\n11.3%
\nBest score per evaluation in bold. Third-party model scores are the best of self-reported or publicly available results.
\nGrok 4.6 is available today in Cursor and Grok Build. It’s also available in the API and other partners like OpenRouter, Vercel, and Cloudflare.
\nPricing starts at $2 per million input tokens and $6 per million output tokens. Additionally, there is a fast variant which is twice the price.
\nWe’re offering 2x included usage inside Grok Build and Cursor for the first week so you can start trying 4.6 immediately.
\nGet started today at x.ai/build.
\n", "date_published": "2026-08-12T00:00:00+00:00", "external_url": "https://x.ai/news/grok-4-6", "id": "urn:aggr:item:6ffec8ba75294911b8a447dd3215a2e092839d93", "image": "https://x.ai/images/news/grok-4-6-og.png", "summary": "Grok 4.6 builds on Grok 4.5 with a particular focus on long-running agents and more ambitious interactive and visual work.", "tags": [ "ai" ], "title": "Introducing Grok 4.6", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-08-12-introducing-grok-4-6/" }, { "content_html": "Today we’re launching Grok Bot, AI teammates you can give real work to.
\nBots have their own computer. They sign into the tools you already use and work across apps, inboxes, and more. They finish jobs end to end, and only come back when something needs your approval.
\nYou can message them like a colleague and hand off the work. They remember conversations, learn how you like things done, and get sharper the more you work together.
\nWe built Grok Bot as an internal prototype, and it took off across the company. Teams created Bots to handle sales outbound, marketing campaigns, office operations, bug fixes, and more. Today we’re excited to open it up to the rest of the world.
\nGrok Bot is in beta and available today for SuperGrok, SuperGrok Plus, and SuperGrok Heavy; Cursor Pro, Pro+, and Ultra; and Cursor Teams Standard and Premium subscribers on desktop and iOS. Grok Bot comes with its own usage, separate from your Grok and Cursor plans, so anything you hand off to a Bot won’t count against your existing usage. Download the app, create your first teammate, and start handing off work. Enterprise users can join a waitlist for future access.
\nEarly access users say Grok Bot feels less like prompting an agent, and more like giving work to a highly capable teammate. Bots share a computer of their own in the cloud, so jobs do not stall when you step away. They can sign in and work across apps, tools, and websites, including platforms with no clean API or MCP, and come back with the work finished.
\nInside SpaceXSI, that already looks like real, multi-step coordinated jobs being handled: a sales Bot updating the CRM with call transcript notes and drafting follow-ups, an ops Bot seating new hires and processing invoices received in Gmail, and an engineering Bot reproducing a bug in the product UI, filing the ticket, and handing the fix off to a debugging Bot.
\n\n\nThere is a huge difference between 90% done and 100% done. Most AI gets you almost there. Grok Bot can finish the swing, because the work lands where a human would put it, in the actual tool.
\n
— Roman, Product
\nIt is easy to get started with Grok Bot. You can message a Bot from your phone or desktop, the way you would text someone on your team, and pick up the same thread later on either surface.
\nOther AI tools may ask you to set up and build workflows and routines first. With Grok Bot, simply message a Bot to take on a task and it gets it done.
\n\n\nThere wasn’t anything to learn. It was just like bringing on a coworker. No automations to set up, no product quirks, no intricate naming. You’re just chatting with a friend.
\n
— Fiona, Community
\nPeople inside SpaceXSI often run multiple Bots in parallel, with one to manage the others. A chief of staff sits on top, with a specialist for each lane: inbox management, expenses, recruiting, bug fixes, or operations. Instead of multiple agents you have to manage, Grok Bot gives you a small team that can work in parallel so you’re not the middleman.
\nBots can independently message each other and share context in threads. When projects overlap, they stay aligned on the same account or project without requiring you to paste notes between chats.
\nYou can also place Bots in a group chat where they can coordinate on their own. They pass work, assign ownership, and only pull you in for judgment calls.
\n\n\nWorking with Grok Bot feels like I have eight arms like an octopus, with every arm in concert with the others, each performing the task the way I would.
\n
— Vincent, Growth
\nThe best way for a Bot to learn your workflow is to ask it to follow along the next time you do a job. It watches the steps and remembers how you like the work done.
\nIt saves your workflow as a routine, takes your corrections, and runs it on its own next time, so you can hand off the same multi-step process without re-explaining it again.
\n\n\nI showed Grok Bot a workflow once and now I just fully trust it to run forever. I feel like I’m 2-3x more efficient because it does it without me verifying and reviewing.
\n
— Bennett, Sales
\nBots are teammates that get sharper over time. They keep context on how you like work done. After a few tasks, they pick up your voice, your edge cases, and know when to ping versus keep going.
\nThey can follow up on threads you dropped, nudge a stalled handoff, and pick work back up from previous conversations.
\nOver time they become more proactive, picking up work before you need to ask and knowing when something needs your attention.
\n\n\nWhen I first started, I was checking in on them every 15 minutes and micromanaging the Bots to the point where they asked me why I kept asking so many questions.
\nNow I let it do its thing and it’s just gotten better with time.
\n
— Emma, Operations
\nA few different ways we have been using Grok Bot — pick a team, and open an example to see the run behind it:
\n", "date_published": "2026-08-11T00:00:00+00:00", "external_url": "https://x.ai/news/introducing-grok-bot", "id": "urn:aggr:item:9b2d4018d3c6594368e7a885fd36221fd51fc031", "image": "https://x.ai/images/news/introducing-grok-bot-og-2.png", "summary": "Grok Bot is your team of always-on agents. They have their own computer, work inside tools and apps like you do, and keep working 24/7.", "tags": [ "ai" ], "title": "Introducing Grok Bot", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-08-11-introducing-grok-bot/" }, { "content_html": "Imagine Image 2.0 is now generally available as the new Quality Mode on grok.com/imagine, and our iOS and Android apps.
\nWe built Image 2.0 around a simple goal: make images you can use in real work. 2.0 follows instructions closely, down to the details. It plans typography and layout the way a designer would, so dense, multi-part visuals hold together and small text comes out sharp. And it preserves what you put in across generations and edits.
\nReal work is iterative. The first generation is rarely the final asset, so Image 2.0 adds tools for changing exactly what you mean and nothing else.
\nThe magic wand edits the region you point at and leaves the rest untouched. Segmentation selects precise areas of the image to change. Background removal exports any subject with a transparent background, ready to drop into other work. Multi-ref editing accepts up to 5 input images in a single generation, removing the need for manual compositing.
\nOne image, any size: pick a ratio and the model fills in the frame.
\nImage 2.0 was trained for fidelity across photography, design, and illustration, with editing treated as a first-class capability. It ranks second in the world in both text-to-image generation and image editing.
\nOverall Elo. Source: Arena Image Edit and Text-to-Image leaderboards (as of Aug 7, 2026). xAI models are listed on Arena under SpaceXAI.
\nWe’re also introducing new templates, which package common image workflows into ready-made starting points. Each one covers a specific task — photo editing, product shots, headshots, icons, game assets, and more — with the workflow already configured, so you supply the inputs and get a finished result.
\nOne world from a handful of prompts: a character, her locations, and the props she carries — generated separately, holding one style from image to image.
\nImage 2.0 is available in the API as grok-imagine-image-2.0.
python
\nFor full documentation, visit our docs page, or start generating in the playground.
\nOur best video model, now with text, image, and voice references — generating up to 1080p.
\n", "date_published": "2026-07-31T00:00:00+00:00", "external_url": "https://x.ai/news/grok-imagine-video-1-5-references", "id": "urn:aggr:item:566db8948dbff316e537810ff5957c92e233897f", "image": "https://media.x.ai/v1/website/grok-imagine-video-1-5-references-og-6190893f.webp", "summary": "Our best video model, now with text, image, and voice references — generating up to 1080p.", "tags": [ "ai" ], "title": "Imagine Video 1.5 with References", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-07-31-imagine-video-1-5-with-references/" }, { "content_html": "Today, we're announcing Grok Voice 2.0, our next-generation voice model with improved intelligence, transcription accuracy, and conversational capabilities.
\nGrok Voice 2.0 is our most intelligent voice model yet, building on its predecessor with meaningful gains in speech reasoning, conversational ability, and tool use reliability.
\nSpeech-to-speech benchmark comparison between Grok Voice 2.0 and competing models, per Artificial Analysis.
\n| Benchmark | \nGrok Voice 2.0 (this release) | \nGrok Voice 1.0 | \nGPT-Realtime-2.1 (High) | \nGemini 3.1 Flash (High) | \n
|---|---|---|---|---|
| 82.9% | \n75.7% | \n79.1% | \n69.5% | \n\n |
| 97.2% | \n97.1% | \n96.0% | \n96.6% | \n\n |
| 95.1% | \n77.8% | \n95.7% | \n74.3% | \n\n |
| 56.5% | \n52.1% | \n45.7% | \n37.7% | \n\n |
| 0.70s | \n1.25s | \n— | \n2.98s | \n\n |
Source: Artificial Analysis.
\nGrok Voice 2.0 outperforms even dedicated, state of the art transcription models when it comes to accuracy. In our evaluation across thousands of short phrases in 24 different languages, we've demonstrated a 1.5–2.0× improvement relative to Deepgram Nova 3 and ElevenLabs Scribe v2, and a 1.4× improvement relative to Grok Voice 1.0.
\nThe gap between Grok Voice 2.0 and dedicated speech-to-text models widens to ~10× in noisy settings. We've focused on making Grok Voice 2.0 perform exceptionally well in real-world settings, with substantial background noise and telephony compression.
\nWord Error Rate (%). Lower is better.
\nGrok Voice 2.0
\nGrok Voice 1.0
\nDeepgram Nova 3
\nElevenLabs Scribe v2
\nGrok Voice models have a unique characteristic: they reason through queries while speaking. Reasoning in parallel with speech makes the model substantially smarter than other speech-to-speech models with no impact on latency.
\nGrok Voice 2.0 has been trained to be very efficient with reasoning tokens relative to its predecessor. In production settings, this means that tool calls are snappier, usually executing before the end of the agent's first sentence.
\nRelative reasoning tokens per response (P50). Lower is better.
\nGrok Voice 2.0
\n0.4×
\nGrok Voice 1.0
\n1.0×
\nWe've trained Grok Voice 2.0 to be a better conversationalist. By using extensive reinforcement learning to push the model towards patterns we see in real human conversation, we've found that broadly the model speaks in shorter sentences, asks one question at a time, and avoids fluff. From the user's perspective, conversations with the agent are simple and fluid, even though the model is often guiding the conversation through complex workflows and thinking several steps ahead behind the scenes.
\nGrok Voice 2.0 is expected to result in improved performance across almost all use cases without any edits to existing prompts. In A/B testing this model on Starlink (+1 888 GO STARLINK), we've seen a significant increase in sales conversion rate and support containment rate.
\nOn August 5, 2026, grok-voice-latest will move from Grok Voice 1.0 to Grok Voice 2.0 (grok-voice-2.0). No action needed to upgrade.
We believe pricing should be predictable and transparent. Grok Voice 2.0 is priced at $0.08 / min of audio.
\nCreate and share websites, apps, games, and interactive dashboards. Now in Early Beta for SuperGrok Heavy subscribers on grok.com, iOS, and Android.
\n", "date_published": "2026-07-28T00:00:00+00:00", "external_url": "https://x.ai/news/grok-build-mode", "id": "urn:aggr:item:9c9b9c6aeaf5f393c64f625fcabf447ae2eb5720", "image": "https://media.x.ai/v1/website/grok-build-mode-og-1454d1ce.png", "summary": "Create and share websites, apps, games, and interactive dashboards. Now in Early Beta for SuperGrok Heavy subscribers on grok.com, iOS, and Android.", "tags": [ "ai" ], "title": "Introducing Build Mode", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-07-28-introducing-build-mode/" }, { "content_html": "Grok 4.5, our smartest coding model, is now available in GitHub Copilot.
\n", "date_published": "2026-07-28T00:00:00+00:00", "external_url": "https://x.ai/news/grok-github-copilot", "id": "urn:aggr:item:fcdca9313c6b2eed2263593d71f27c8d3164e5ad", "image": "https://media.x.ai/v1/website/og-github-copilot-603406c4.webp", "summary": "Grok 4.5, our smartest coding model, is now available in GitHub Copilot.", "tags": [ "ai" ], "title": "Grok 4.5 in GitHub Copilot", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-07-28-grok-4-5-in-github-copilot/" }, { "content_html": "Use the Grok add-on for Google Workspace to work directly in Sheets, Slides, and Docs.
\n", "date_published": "2026-07-24T00:00:00+00:00", "external_url": "https://x.ai/news/introducing-google-workspace-addon", "id": "urn:aggr:item:d7bf15540f7bd8a3edc41257c80c327f37afd806", "image": "https://media.x.ai/v1/website/grok-in-google-workspace-og-129ad993.png", "summary": "Use the Grok add-on for Google Workspace to work directly in Sheets, Slides, and Docs.", "tags": [ "ai" ], "title": "Grok in Google Workspace", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-07-24-grok-in-google-workspace/" }, { "content_html": "Grok Build can now run workflows: describe a large task in plain language, and Grok plans it, fans it out across hundreds of parallel agents in the background, and reports back when everything is done. Your session stays free the whole time.
\nWorkflows are built for complex, multi-faceted tasks a single conversation can't hold: reviewing every feature in a large PR, triaging the last 100 issues, auditing a codebase for one class of bug. If the work splits into many independent pieces and should end in one clear report, ask for a workflow.
\nGrok plans the task as a small script: the phases of work, the agents in each phase, and how their results roll up. Each agent starts with a clean, focused context, and the plan can build in checks a single pass can't, like independent skeptics verifying every finding before it reaches the final report.
\nRuns get a budget of 128 agents, and up to 1,024 for big jobs. Progress is saved as the run goes, so pausing and resuming never redoes finished work. Run /workflows to watch it live, phase by phase, with per-agent token counts.
Grok authors the workflow from your request, smoke-checks it before launching, and improves it between runs; you never write the script yourself. When one works, keep it: workflows in .grok/workflows/ are shared with your team, and ones in ~/.grok/workflows/ follow you everywhere. Each saved workflow becomes its own slash command that takes arguments, so once you keep the PR review from the demo above, the next review is just /pr-review 5137.
/deep-research ships built in: it fans research questions out to parallel investigators, verifies every claim against its sources, and returns a cited report.
Some examples to try:
\nGrok 4.5, our most intelligent model yet, is now on grok.com, X, iOS, and Android.
\n", "date_published": "2026-07-22T00:00:00+00:00", "external_url": "https://x.ai/news/grok-4-5-everywhere", "id": "urn:aggr:item:7126807f520ccff7ed85af8554a7dc69e93a2b6e", "image": "https://media.x.ai/v1/website/grok-4-5-everywhere-og-e052c7ec.png", "summary": "Grok 4.5, our most intelligent model yet, is now on grok.com, X, iOS, and Android.", "tags": [ "ai" ], "title": "Bringing Grok 4.5 to iOS, Android, Web, and X", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-07-22-bringing-grok-4-5-to-ios-android-web-and-x/" }, { "content_html": "Use the Grok add-in for Microsoft Outlook to summarize long email threads, draft replies in your voice, and keep your inbox organized.
\n", "date_published": "2026-07-21T00:00:00+00:00", "external_url": "https://x.ai/news/introducing-outlook-addin", "id": "urn:aggr:item:7bab6523682fc552974eb28a5964439a875e36da", "image": "https://media.x.ai/v1/website/introducing-outlook-addin-og-44242ed8.jpg", "summary": "Use the Grok add-in for Microsoft Outlook to summarize long email threads, draft replies in your voice, and keep your inbox organized.", "tags": [ "ai" ], "title": "Introducing Grok for Outlook", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-07-21-introducing-grok-for-outlook/" }, { "content_html": "Use the Grok add-in for Microsoft Excel to ask questions in plain English, write formulas, and run scenarios without leaving the workbook.
\n", "date_published": "2026-07-20T00:00:00+00:00", "external_url": "https://x.ai/news/introducing-excel-addin", "id": "urn:aggr:item:34b6c2114d06ec66639b4112b625b240102d7ef8", "image": "https://media.x.ai/v1/website/introducing-excel-addin-title-746bb775.webp", "summary": "Use the Grok add-in for Microsoft Excel to ask questions in plain English, write formulas, and run scenarios without leaving the workbook.", "tags": [ "ai" ], "title": "Introducing Grok for Excel", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-07-20-introducing-grok-for-excel/" }, { "content_html": "Today, we're launching Grok 4.5, SpaceXSI's smartest model built to excel at coding, agentic tasks, and knowledge work. It's our strongest model ever and was trained alongside Cursor.
\nGrok 4.5 was trained on datasets spanning knowledge in coding, science, engineering, and math. With both intelligent and efficient reasoning, Grok 4.5 excels at real engineering tasks and exceeds comparable leading models at these tasks.
\nCompetitor figures are drawn from the respective developers’ published system cards or benchmark leaderboards
\nBenchmark bar charts of model scores. DeepSWE 1.0 (eval created by Datacurve, run with each model provider's harnesses by AA): Fable (max) 66.1%, GPT 5.5 (xhigh) 64.31%, Grok 4.5 62.0%, Opus 4.8 (max) 55.75%, Opus 4.7 (max) 40.12%. DeepSWE 1.1 (mini-swe-agent harness run by Datacurve): Fable (max) 70%, GPT 5.5 (xhigh) 67%, Opus 4.8 (max) 59%, Grok 4.5 53%, GLM 5.2 44%. SWE Marathon resolution rate (pass@1): Grok 4.5 29.0%, Opus 4.8 (max) 26.0%, Fable (max) 24.0%, Opus 4.7 (max) 16.0%. Terminal Bench 2.1: Fable (max) 84.3%, GPT 5.5 (xhigh) 83.4%, Grok 4.5 83.3%, Opus 4.8 (max) 78.9%, Opus 4.7 (max) 78.9%. SWE Bench Pro resolve rate: Fable (max) 80.4%, Opus 4.8 (max) 69.2%, Grok 4.5 64.7%, Opus 4.7 (max) 64.3%, GLM 5.2 62.1%, GPT 5.5 (xhigh) 58.6%. Competitor figures are drawn from the respective developers' published system cards or benchmark leaderboards.
\nGrok 4.5 was trained across tens of thousands of NVIDIA GB300 GPUs, with training and stability techniques designed for large-scale runs. Beyond raw token volume, we invested heavily in data filtering and curation: deduplication, quality scoring, and domain-focused selection so that the data mixture stayed high-coverage and high-signal.
\nWe scaled reinforcement learning with a strong focus on per-token intelligence. Our RL training covers hundreds of thousands of tasks, centered on multi-step software engineering and other technical work, with automated and model-based grading. Our stack is built for highly asynchronous training, so agentic rollouts can run for many hours while learning continues across tens of thousands of GPUs. The result is more intelligent and efficient reasoning on real engineering and agentic tasks.
\nGrok 4.5 is incredibly capable at coding, from challenging Rust and C/C++ tasks to end-to-end app building from prompt to production. Below are some examples built by the model with one prompt. Grok 4.5 is highly proficient at creating well-designed, end-to-end functional apps even with minimal specification.
\nMake a beautiful simulation of the universe and solar system. should be sped up with adjustable time, realistic motion, orbits, stars. use threejs. Make the HUD well styled and conform to modern design principles.
\nGrok 4.5 is served at fast-model speeds of 80 TPS. Combined with twice greater token efficiency than the latest leading models at the same tasks, the model delivers intelligent results to you more quickly and at far lower costs.
\navg. output tokens per SWE Bench Pro task
\nGrok 4.5 0
\nOpus 4.8 (max) 0
\n4.2× fewer tokens
\n0 70k tokens
\nToken efficiency, average output tokens per SWE Bench Pro task — Grok 4.5 resolves tasks with 15,954 output tokens on average, about 4.2× fewer than Opus 4.8 (max) at 67,020
\nGrok 4.5 is now the default model in Grok Build. In addition to its coding proficiency, Grok Build is capable of building complex Excel models that involve research from the web, multi-sheet formula use, and even leaves stickies or notes behind for future reference.
\nIn PowerPoint and Word, Grok 4.5 is similarly meticulous. The model is capable of using native PowerPoint shapes to build complex diagrams, designing intuitive slide content, and writing clear prose in Word.
\nOutline a 5-slide quarterly business review
\nLearn more about our plugins for Word, PowerPoint, Excel, and Outlook.
\nGrok 4.5 is delivered at an incredibly competitive cost compared to other leading models. Grok 4.5 is priced at $2 per million input tokens and $6 per million output tokens. The model also achieves roughly 2x the token efficiency of comparable leading models, solving tasks in under half the number of steps. Overall, Grok 4.5 delivers the highest intelligence per unit of time and cost.
\nGrok 4.5 is available today in Grok Build, in Cursor on all plans, and from the SpaceXSI console. Simply grab an API key and get started in a few lines of code:
\nbash
\nWe’re offering free Grok 4.5 usage for a limited time in Grok Build and Cursor. Get started today at x.ai/build.
\n", "date_published": "2026-07-16T00:00:00+00:00", "external_url": "https://x.ai/news/grok-4-5", "id": "urn:aggr:item:436844980deae3a51b5638f26e36f2b71d310c59", "image": "https://media.x.ai/v1/website/grok-4-5-og-786d2a81.png", "summary": "Grok 4.5 is SpaceXSI's smartest model built for coding, agentic tasks, and knowledge work.", "tags": [ "ai" ], "title": "Introducing Grok 4.5", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-07-16-introducing-grok-4-5/" }, { "content_html": "Today, we're introducing Automations: jobs Grok runs on its own. Describe the work once, choose when it runs, and Grok takes it from there, whether that's research done before you're awake or an important email flagged the moment it lands. Automations are available on grok.com and in the Grok app on iOS and Android.
\nAn automation's instructions read like any chat message: describe what you want, attach files for context, add connectors and skills, and pick a mode. Name it, save it, and from then on every run is a fresh request: same instructions, current data.
\nSchedules run once, daily, weekdays, weekly, monthly, or yearly, at a time you choose in your timezone: a morning brief at 8:00 before the day starts, or a rent reminder on the 1st.
\nEmail triggers watch your inbox instead. When an incoming email matches your filters (sender, recipient, or subject), the automation fires with that email as context, and Grok responds to the actual message. Instructions can also point at your tools directly: type @ to mention a connector, and Grok uses it on every run.
\nYou can also start a run yourself with Run now, which is handy for testing an automation right after you build it.
\nWhen an automation fires, Grok opens a real conversation, does the work, and saves the result to its run history. Open any run to read the full thread, or pick up the conversation where Grok left off. You choose how each automation reports back: email, app notification, both, or neither if you'd rather check in yourself.
\nYou can also create automations straight from chat: ask Grok to "check the news every morning and flag anything about pricing" and it sets one up for you. The Automations page has suggested templates to start from, and any automation can be paused, resumed, edited, or deleted at any time.
\nCreate your first automation at grok.com/automations. Scheduled automations are available to everyone; email triggers are included with SuperGrok.
\n", "date_published": "2026-07-16T00:00:00+00:00", "external_url": "https://x.ai/news/grok-automations", "id": "urn:aggr:item:6270aaa40f9e79a68f41f6744eb8c9cad159e536", "image": "https://media.x.ai/v1/website/grok-automations-og-93199cf4.jpg", "summary": "Describe a job once and Grok runs it on a schedule or when an email arrives, then reports back.", "tags": [ "ai" ], "title": "Automations in Grok", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-07-16-automations-in-grok/" }, { "content_html": "Explore the harness behind our coding agent and TUI.
\n", "date_published": "2026-07-15T00:00:00+00:00", "external_url": "https://x.ai/news/grok-build-open-source", "id": "urn:aggr:item:158b51e1c8f82ca7cfc0c96fe7ac75519b41af25", "image": "https://media.x.ai/v1/website/open-source-cover-930d1067.png", "summary": "Explore the harness behind our coding agent and TUI.", "tags": [ "ai" ], "title": "Grok Build is Now Open Source", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-07-15-grok-build-is-now-open-source/" }, { "content_html": "New multilingual voices for Grok Voice, plus improved naturalness for the original five.
\n", "date_published": "2026-07-06T00:00:00+00:00", "external_url": "https://x.ai/news/new-flagship-voices", "id": "urn:aggr:item:5ef32c07f31531eb92c6be9b9a05b9a4a1e0ab6a", "image": "https://media.x.ai/v1/website/new-flagship-voices-6e40ffad.webp", "summary": "New multilingual voices for Grok Voice, plus improved naturalness for the original five.", "tags": [ "ai" ], "title": "21 New Flagship Grok Voices", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-07-06-21-new-flagship-grok-voices/" }, { "content_html": "Today, we’re announcing Voice Agent Builder in beta: a no-code platform to configure production voice agents on Grok Voice.
\nIt’s for operators and developers who want high-volume production voice agents without building the surrounding stack from scratch. Out of the box you get telephony, knowledge retrieval, tools, guardrails, MCPs, and observability in one place. You can also keep what you already have: bring existing phone numbers over SIP, wire tools to your APIs and MCP servers, or connect your own client over WebSocket.
\nMost voice stacks stitch together three APIs—speech-to-text, a language model, and text-to-speech—often with each stage hosted by a different provider. Every hop adds cost, latency, and new failure modes. Voice Agent Builder is one interface on a speech-to-speech path built for Grok Voice, tightly coupled to the model rather than assembled from three.
\nReal calls come with low-quality telephony audio, background noise, strong accents, interruptions, and callers who change their minds mid-sentence. The workflows behind them are ambiguous, run across dozens of tools, and happen in any of 25+ languages.
\nWe trained Grok Voice on those calls. τ-voice Bench measures agents under the same conditions.
\nτ-voice Bench Leaderboard
\nGrok Voice 1.0
\n67.3%
\nGemini 3.1 Flash Live
\n43.8%
\nGPT Realtime 1.5
\n35.3%
\nSetup is simple: write a plain-language description of how calls should flow, then attach your documents, tools, and guardrails. You can go from zero to a working agent in about two minutes.
\nAn agent starts with a prompt that describes how calls should go. The model reasons in real time, so it can follow long instructions and work through ambiguous requests.
\nWhat it knows comes from the knowledge base. You upload documents in common formats (plain text, Markdown, Word, PowerPoint, Excel, HTML, JSON, and others), and the agent retrieves from them during calls. Documents are organized into collections, which you can attach to one or more agents and share across agents so policies, product specs, and runbooks stay in one place instead of being pasted into every prompt.
\nKnowing the business is only half of a support or sales call. Agents also need to act. They look things up, change records, hand off, or close the loop after the conversation.
\nTools and connectors are how that happens. On a booking line, the agent might schedule appointments in Google Calendar or Outlook Calendar, then send a confirmation through your email provider. On support, an API request can check order status or issue a refund in your own systems. When the answer isn't only in your documents, web search or X search can pull current public information. Tickets can be managed in Linear or Notion, and files come from Google Drive or OneDrive.
\nIf the caller needs a human, the agent can transfer the call to your team. When the task is complete, it can end the call cleanly. Throughout the conversation, it sends real-time notifications so your team can see what the agent did and step in if needed.
\nAgents can use any of the built-in voices, or a clone of your brand's voice made from about two minutes of audio. Each account includes a free phone number, ready for anything from a first test call to production traffic, and direct SIP connects an existing number from any major telephony provider. You can also test changes in the browser without a phone.
\nEvery call is recorded and transcribed. You can play back the audio, read the transcript, and see which tools the agent used. Guardrails set limits on what the agent shouldn't do, like reading back card numbers or discussing topics off script.
\nWe believe that pricing should be simple and transparent. Agents are billed at our API rate (currently $0.08 / min of audio), with voices included and no separate platform fee. Telephony on a free provisioned number is an additional $0.01 / min.
\nOther voice stacks commonly bill for each individual component (recognition, reasoning, synthesis, and platform), each with its own meter and pricing. We wanted a small number of meters you can multiply by call volume and be done.
\nA voice agent is easier to judge by ear than by benchmark. Build one, give it your hardest workflow, and call it.
\n", "date_published": "2026-07-01T00:00:00+00:00", "external_url": "https://x.ai/news/grok-voice-agent-builder", "id": "urn:aggr:item:50ef683c1d95c5c0148931a29437aa1041f5e050", "image": "https://media.x.ai/v1/website/grok-voice-agent-builder-0caac71c.webp", "summary": "Create a personalized voice agent in under 2 minutes without a single line of code.", "tags": [ "ai" ], "title": "Introducing the Voice Agent Builder", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-07-01-introducing-the-voice-agent-builder/" }, { "content_html": "Use /goal for long-running autonomous task execution in Grok Build.
\n", "date_published": "2026-06-22T00:00:00+00:00", "external_url": "https://x.ai/news/introducing-goal", "id": "urn:aggr:item:994d913aa3201cd1152f6f7a50324a1eb06ac880", "image": "https://media.x.ai/v1/website/introducing-goal-og-62e9a5f8.webp", "summary": "Use /goal for long-running autonomous task execution in Grok Build.", "tags": [ "ai" ], "title": "Introducing /goal", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-06-22-introducing-goal/" }, { "content_html": "Use the Grok add-in for Microsoft Word to turn notes into documents, style and format your work, or bring research from the web into Word.
\n", "date_published": "2026-06-18T00:00:00+00:00", "external_url": "https://x.ai/news/introducing-word-addin", "id": "urn:aggr:item:676202fbc8040e5e61b3624db63631fe3cbfd846", "image": "https://media.x.ai/v1/website/introducing-word-addin-title-abf047af.webp", "summary": "Use the Grok add-in for Microsoft Word to turn notes into documents, style and format your work, or bring research from the web into Word.", "tags": [ "ai" ], "title": "Introducing Grok for Word", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-06-18-introducing-grok-for-word/" }, { "content_html": "Grok models are now available on Databricks Agent Bricks.
\n", "date_published": "2026-06-18T00:00:00+00:00", "external_url": "https://x.ai/news/grok-databricks", "id": "urn:aggr:item:4e34e04dc195d652ceae00278a8616ed28e214d5", "image": "https://media.x.ai/v1/website/grok-databricks-21d5922f.webp", "summary": "Grok models are now available on Databricks Agent Bricks.", "tags": [ "ai" ], "title": "Grok on Databricks", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-06-18-grok-on-databricks/" }, { "content_html": "Grok models are now available via Amazon Bedrock.
\n", "date_published": "2026-06-17T00:00:00+00:00", "external_url": "https://x.ai/news/grok-amazon-bedrock", "id": "urn:aggr:item:5ca08c3c773d5d5a5579e8dbc38c86dc9fed4edf", "image": "https://media.x.ai/v1/website/grok-aws-bedrock-e15961af.jpg", "summary": "Grok models are now available via Amazon Bedrock.", "tags": [ "ai" ], "title": "Grok on Amazon Bedrock", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-06-17-grok-on-amazon-bedrock/" }, { "content_html": "Grok now works inside Microsoft PowerPoint — turn outlines into slides, expand the deck, and tighten the narrative without leaving the app.
\n", "date_published": "2026-06-16T00:00:00+00:00", "external_url": "https://x.ai/news/introducing-powerpoint-addin", "id": "urn:aggr:item:30c6f74ad33fd6114a3ef1db5dd88691c9474c09", "image": "https://media.x.ai/v1/website/grok-ppt-3c8fe7c3.webp", "summary": "Grok now works inside Microsoft PowerPoint — turn outlines into slides, expand the deck, and tighten the narrative without leaving the app.", "tags": [ "ai" ], "title": "Introducing Grok for PowerPoint", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-06-16-introducing-grok-for-powerpoint/" }, { "content_html": "Grok Imagine Video 1.5 is now generally available on the Imagine API. We've also rolled out Video 1.5 Fast on grok.com/imagine and our iOS and Android apps. These are our best image-to-video models yet: better motion, better physics, better audio, at the fastest speeds. We’re also adding features to enhance your creative workflow in Grok Imagine.
\nCompared to the previous model, 1.5 improves across every dimension that matters for real creative work:
\nSound effects, ambience, and dialogue are generated in the same pass and land on the action. Speech is clearer and better synced.
\nMovement holds together over the length of a clip — fewer warps, more believable weight and momentum.
\nGrok Imagine Video 1.5 Fast almost doubles generation speed: it produces 6-second, 720p videos in about 25 seconds, down from 40+ seconds in our previous model.
\nTake a look behind the scenes at how David Thompson (@heavypulp) made a trailer worthy of the big screen entirely with Grok Imagine 1.5:
\nAlongside Imagine Video 1.5, we're rolling out new features over the next few days to improve your creative workflow:
\nOrganize your work into Projects that appear on the left sidebar.
\nKick off multiple agents in parallel on your projects. Instead of waiting for one generation to finish before starting the next, you can run several prompts at once and let them work for you.
\nFind any image or video you've made by searching your library — no more scrolling to track down that one clip.
\nImagine Video 1.5 is out of preview and generally available in the xAI API as grok-imagine-video-1.5. Give it a starting image, describe the motion, and choose your resolution and duration.
python
\nWe can't wait to see what you make.
\n", "date_published": "2026-06-16T00:00:00+00:00", "external_url": "https://x.ai/news/grok-imagine-video-1-5", "id": "urn:aggr:item:bdd763d87e1568e6d45ebdd73f5087842202314e", "image": "https://media.x.ai/v1/website/grok-imagine-video-1-5-og-10a5c726.png", "summary": "Improved quality at even faster speeds.", "tags": [ "ai" ], "title": "Grok Imagine Video 1.5", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-06-16-grok-imagine-video-1-5/" }, { "content_html": "Use your Grok and X Premium subscription inside Warp.
\n", "date_published": "2026-06-15T00:00:00+00:00", "external_url": "https://x.ai/news/grok-warp", "id": "urn:aggr:item:5da1f5529ebeff8d438805c83d910a3a22a88c22", "image": "https://media.x.ai/v1/website/grok-warp-912f684e.webp", "summary": "Use your Grok and X Premium subscription inside Warp.", "tags": [ "ai" ], "title": "Use Grok in Warp", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-06-15-use-grok-in-warp/" }, { "content_html": "The Agent Dashboard puts every Grok Build session on one screen. See what each is doing, run them in parallel, and step in only when input is needed.
\nRun grok dashboard from your shell, or /dashboard (Ctrl+\\) from inside any session.
The dashboard sorts sessions by state, with anything waiting for input pulled to the top, so you handle blockers first and leave the rest running. A quick scan shows what each session is doing and for how long, so you stay oriented without opening anything.
\nSpread across repos? Group by working directory with Ctrl+S. Subagents roll up under the session that launched them, so the list shows the work you dispatched, not the fan-out beneath it.
Select a row to peek at its latest output without leaving the dashboard, then reply from there. Idle sessions send immediately; active ones queue your message until the current turn ends.
\nWhen a session requests approval or asks a question, its options appear inline. Answer with the arrow or number keys to continue. Multi-part questions arrive one at a time.
\nThe input at the bottom starts a new session. Enter dispatches it and keeps you on the dashboard; Shift+Enter dispatches and opens it right away. Before sending, set the model, start in plan mode, or let the session approve its own edits.
Open any session to take over its full conversation. Cycle to the next or previous session without returning to the list, then drop back to the dashboard when you're done. Closing the dashboard leaves every session running, and they're all there when you reopen it.
\nThe Agent Dashboard ships with Grok Build. Install it with a single command, then run grok dashboard, or /dashboard from any session you already have open.
Launching the built-in plugin marketplace for Grok Build.
\n", "date_published": "2026-06-11T00:00:00+00:00", "external_url": "https://x.ai/news/grok-plugin-marketplace", "id": "urn:aggr:item:d3d166d15ef7974a7a7c4a9c59f54f4e794fc9f3", "image": "https://media.x.ai/v1/website/grok-plugin-marketplace-937ebfa4.webp", "summary": "Launching the built-in plugin marketplace for Grok Build.", "tags": [ "ai" ], "title": "Grok Build Plugin Marketplace", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-06-11-grok-build-plugin-marketplace/" }, { "content_html": "Our latest image-to-video model, grok-imagine-video-1.5-preview, is now available via the xAI API in preview.
grok-imagine-video-1.5-preview turns a single still image into fluid, cinematic video. Give it a starting frame and a prompt describing the motion, and it animates the scene, including camera moves, atmosphere, and physics, while staying faithful to your source image. You can generate clips at up to 720p.
Direct the shot with natural-language prompts. Describe the camera move, the pacing, and the sound design, then set your resolution and clip length. The model holds detail and lighting from the input frame, so the result continues the original image rather than reinterpreting it.
\nThe model also works well for sequences. Stage each frame, animate it, and chain the shots together into longer scenes that keep a consistent look across an entire project.
\nAnimate an image in a few lines of code.
\npython
\n", "date_published": "2026-06-03T00:00:00+00:00", "external_url": "https://x.ai/news/grok-imagine-1-5", "id": "urn:aggr:item:e65c24c830850b1b07d249353a9c09f883d89a9a", "image": "https://media.x.ai/v1/website/grok-imagine-1-5-og-342e9fa3.webp", "summary": "grok-imagine-video-1.5-preview, our latest image-to-video model, is now available via the xAI API in preview.", "tags": [ "ai" ], "title": "Grok Imagine 1.5 Preview", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-06-03-grok-imagine-1-5-preview/" }, { "content_html": "Composer 2.5 is now available in Grok Build. Try it from the /models menu.
\n", "date_published": "2026-06-01T00:00:00+00:00", "external_url": "https://x.ai/news/composer-2-5", "id": "urn:aggr:item:958244291c594b0f84ce55484e92203fc4860473", "image": "https://media.x.ai/v1/website/composer-2-5-og-a4729aa0.png", "summary": "Composer 2.5 is now available in Grok Build. Try it from the /models menu.", "tags": [ "ai" ], "title": "Composer 2.5", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-06-01-composer-2-5/" }, { "content_html": "Our latest coding model, grok-build-0.1, is now available via the xAI API in public beta.
grok-build-0.1 is a coding model specifically trained for agentic coding tasks, including web development, debugging, and MCP support. It's the same model that powers Grok Build.
The model is served at a blazing-fast 100+ tokens/second and is priced at $1/m tokens in and $2/m tokens out. Outside of coding, it’s also a speedy, economical option for general-purpose agentic and tool calling use cases.
\nbash
\ngrok-build-0.1 performs best in agentic harness like Grok Build, Cursor, Hermes Agent, OpenClaw, Kilo Code, or OpenCode. It’s also available via OpenRouter and Vercel AI Gateway.
Use your SuperGrok or X Premium+ subscription inside Kilo Code, the open-source agentic coding platform.
\n", "date_published": "2026-05-27T00:00:00+00:00", "external_url": "https://x.ai/news/grok-kilocode", "id": "urn:aggr:item:7a9c0137a69d52134c001eb446a810b74db934f5", "image": "https://media.x.ai/v1/website/grok-kilocode-4e041462.png", "summary": "Use your SuperGrok or X Premium+ subscription inside Kilo Code, the open-source agentic coding platform.", "tags": [ "ai" ], "title": "Use Grok in Kilo Code", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-05-27-use-grok-in-kilo-code/" }, { "content_html": "Now in early beta for all SuperGrok and X Premium Plus subscribers — Grok Build is a new coding agent that runs right from your terminal.
\n", "date_published": "2026-05-25T00:00:00+00:00", "external_url": "https://x.ai/news/grok-build-cli", "id": "urn:aggr:item:0ec13fbb78c576bf7e8e231a84ad6c0ad6eb8668", "image": "https://media.x.ai/v1/website/grok-build-cli-og-5fe6e581.png", "summary": "Now in early beta for all SuperGrok and X Premium Plus subscribers — Grok Build is a new coding agent that runs right from your terminal.", "tags": [ "ai" ], "title": "Introducing Grok Build", "url": "https://aggr.aymericbeaumet.com/items/x-ai/2026-05-25-introducing-grok-build/" } ], "language": "en", "title": "News: Research, Product & Company Updates | SpaceXSI", "version": "https://jsonfeed.org/version/1.1" }