Grok 4.6 Explained: What's New and Whether You Get It

SpaceXAI shipped a new flagship today, and the honest headline for most people is that it is not for you yet. On August 12, 2026, the company released Grok 4.6, describing it as a model built for "long-running agents and more ambitious interactive and visual work." It is a real step up on SpaceXAI's own benchmarks. It is also, today, a developer release: the places SpaceXAI says it runs are all tools you write code in, and the Grok app on your phone is still running Grok 4.5.

Key Takeaways

• Grok 4.6 launched August 12, 2026 as SpaceXAI's frontier model for coding, agents and knowledge work.
• It is not in the Grok app. SpaceXAI's model page lists where it runs -- the xAI API, Grok Build, Cursor, and the OpenRouter, Vercel and Cloudflare gateways. Neither grok.com nor the phone app is on that list.
• We checked the app directly. On launch day, grok.com's own mode picker still described Fast, Expert and Heavy as running Grok 4.5, with Fast as the default.
• So nothing changes for you today. No download, no setting, no new bill. Free Grok is still free and still Grok 4.5.
• It matches GPT-5.6 Sol rather than beating it. SpaceXAI scores Grok 4.6 at 61 on the Artificial Analysis Intelligence Index against Sol's 61 and Claude Fable 5's 62.
• It is cheap for developers: $2 per million input tokens and $6 per million output, the same headline rate Grok 4.5 launched at, with a 500,000-token context window.
• The free trial is for coders, not chatters. SpaceXAI is giving 2x included usage in Grok Build and Cursor for the first week.
• Grok 4.6 is already the default in Grok Build, SpaceXAI's coding agent -- which is the clearest signal of who this release is actually for.

Do You Get It? The One-Line Answer by How You Use Grok

If you use Grok the way most people do -- the app, the website, or the Grok button on X -- you do not get it today. Here is the full picture, taken from SpaceXAI's own launch post and model documentation on August 12, 2026, the day of release.

The line that matters is the one SpaceXAI wrote itself. Its model page carries a heading called "Where it runs," and it names four things: the xAI API, Grok Build, Cursor, and model gateways. A vendor listing its surfaces and leaving its own consumer app off the list is a much stronger signal than silence -- and grok.com backs it up, because the page still describes every one of its four modes as running Grok 4.5.

What Grok 4.6 Actually Does Better

It sticks with long jobs. SpaceXAI's framing is that Grok 4.6 "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." The company also says it saw "more self-testing and verification, with the model checking its own work before moving on" on longer runs.

In plain terms, this is not a model built to answer your question better. It is a model built to be left alone with a task for an hour. That is why it landed in a coding agent and a code editor rather than in a chat app, and it is the honest reason the release does not change your day.

The second claim is about first drafts: SpaceXAI says Grok 4.6 "produces stronger first passes on visual and interactive projects" and can "establish structure and visual language for an application in one pass." Again, useful if you build things; invisible if you ask questions.

The Benchmarks, and What They Are Worth

Grok 4.6 draws level with the best rather than passing them, on SpaceXAI's own scoreboard. The company published a ten-row comparison table against Grok 4.5, GPT-5.6 Sol and Claude Fable 5. Here are the rows an outsider can most easily interpret.

Two things stand out, and both come straight from SpaceXAI's own numbers. The jump over Grok 4.5 is large and consistent -- it wins all ten rows in the full table. But against the competition it is a tie at best: it matches GPT-5.6 Sol on the headline index, sits just under Claude Fable 5, and loses badly on two coding tests. SpaceXAI says competitor figures are "drawn from the respective developers' published system cards or benchmark leaderboards," which is the fairest available method and still means the company scored its own exam. Treat any "Grok beats ChatGPT" headline you see today with the suspicion it has earned.

What It Costs

Cheap, and unchanged from Grok 4.5's headline rate. Per SpaceXAI's release notes, Grok 4.6 costs $2 per million input tokens, $0.50 per million cached input tokens, and $6 per million output tokens below 200,000 prompt tokens, doubling to $4 / $1 / $12 above that. The launch post adds that "there is a fast variant which is twice the price."

None of this reaches a consumer bill, because there is no consumer product to bill it against yet. If you are weighing what to actually pay for a chatbot this month, our guide to which AI chatbot is worth paying for covers the plans that exist today, and the best free AI chatbot in 2026 covers what you can get without paying.

Should You Care?

Ignore it if you use the Grok app. There is nothing to enable and nothing to wait for on a published date. Open Grok, leave it on Fast, and you are on Grok 4.5 -- still free, still the model our Grok 4.5 explainer walks through plan by plan.

Do not upgrade to SuperGrok because of this. A paid Grok plan buys you Expert and Heavy modes and higher limits. As of today it does not buy you Grok 4.6, because no consumer plan has it.

Care if you write code or build agents. This is the release's real audience. Cursor has it on every plan, Grok Build defaults to it, and the first week comes with double the included usage.

Watch for one thing: the moment grok.com's mode picker starts saying "Grok 4.6" behind Fast, the free tier gets a frontier model at no cost -- which is exactly what happened with Grok 4.5 in July. SpaceXAI has published no date for that, and we will update this page when it does.

The bottom line: Grok 4.6 is a genuine step up from Grok 4.5 and a genuine tie with the best models from OpenAI and Anthropic, judged by the only scoreboard that exists on day one -- SpaceXAI's own. It is also a release aimed squarely at developers, and the company's own list of where it runs leaves out the app most people mean when they say Grok. If that is you, today's answer is: nothing changed, and you are not missing anything you could have bought.