Grok 4.6 Explained: Free on Every Plan, and Grok 4.7 Has Taken the Paid Modes

SpaceXAI shipped a new flagship on August 12, 2026 and, for the first days, the honest headline for most people was that it was not for them. That has changed. On August 12 the company released Grok 4.6, describing it as a model built for "long-running agents and more ambitious interactive and visual work," and pointed it at developer surfaces only. Since then it has quietly replaced Grok 4.5 across SpaceXAI's entire consumer plan lineup, free plan included, without an announcement.

Key Takeaways

• Grok 4.6 launched August 12, 2026 as SpaceXAI's frontier model for coding, agents and knowledge work, on developer surfaces only.
• It is now on every consumer plan, Free included. SpaceXAI's pricing page ticks a single Grok 4.6 row across all seven tiers it lists, and the Grok 4.5 row is gone.
• You do not have to pay for it. Free is one of the ticked tiers. What a subscription buys is reasoning depth and limits, not a better model.
• The app has moved past it. Read October 2, 2026, grok.com's mode picker names Grok 4.7 behind the paid Expert and Heavy modes, not Grok 4.6. It still names no model behind the free Fast mode, which is still the default.
• Grok 4.7 arrived September 21, 2026 and the pricing page has not acknowledged it. Seven days after the announcement, the September 28 capture of that page still shows one model row reading Grok 4.6, ticked for all seven plans. SpaceXAI's Grok 4.7 post names its availability as Cursor, Grok Build, the API, coding harnesses and cloud routers, and does not mention the consumer apps.
• SpaceXAI never announced the consumer rollout. The only proof is the pricing page itself, which changed between August 12 and August 15, 2026.
• 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.
• Its developer docs still have not caught up. The "Where it runs" list on SpaceXAI's own model page names the API, Grok Build, Cursor and three gateways, and no consumer app, as of its September 14, 2026 update.

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

Yes, whatever you pay. SpaceXAI's plan comparison matrix lists exactly one model row and ticks it for every tier, so the answer for a normal person is that the newest Grok is the Grok you already have. Here is the full picture, read off SpaceXAI's own pricing page (newest readable copy, September 5, 2026) and grok.com's mode picker on September 16, 2026.

Two rows deserve the caveat they carry. SpaceXAI's pricing matrix makes a positive statement that Free includes Grok 4.6, and decoded cell by cell on the September 28, 2026 capture it still does -- the only row the Free column does not get is Video generation. But the app's picker has since relabelled both paid modes to Grok 4.7, so the two surfaces now name different models rather than describing the same one at different levels of detail. The free Fast mode remains unlabelled on both, which means no SpaceXAI surface states which model the free plan runs. Where they disagree on plan inclusion we still follow the pricing page, because that is the surface that states inclusion outright; where they disagree on which model a mode runs, the picker is the one the app actually obeys.

What You Actually Pay For

Not the model. SpaceXAI publishes three consumer prices in the text of its pricing page: Free at $0, SuperGrok at $30 a month, and SuperGrok Plus at $100 a month, with Lite and Heavy tiers alongside them whose prices the page renders too late to read reliably. What the matrix shows the free plan missing is capability, not intelligence.

That is the same shape the free plan had under Grok 4.5, which is the point: SpaceXAI has changed the model on the free tier twice now without changing what the free tier costs. If you are deciding where your subscription money goes 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.

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 explains why it landed in a coding agent and a code editor first, and why the consumer rollout happened without a launch post: from SpaceXAI's point of view, nothing about your day was supposed to change.

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; hard to notice 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, which is the practical case for the swap on the free tier. 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 with the suspicion it has earned.

What It Costs Developers

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."

Should You Care?

Open the app and you already have it. There is nothing to enable, download or upgrade. The pricing page puts Grok 4.6 on the free plan, and the model picker's default mode is still Fast.

Do not upgrade to SuperGrok for the model. A paid Grok plan buys Expert and Heavy reasoning, video generation, unrestricted real-time search and higher limits. The model underneath is the same one the free plan gets, which is exactly what our Grok 4.5 explainer found in July and why our verdict on paying for Grok has not moved.

Care most if you write code or build agents. This is still the release's real audience. Cursor has it on every plan and Grok Build defaults to it.

Treat SpaceXAI's silence as normal, not as a signal. This is the second time the company has put a new flagship on its free tier without a post announcing it, and the second time its developer documentation lagged its own pricing page by more than a week. If you want to know which model you are talking to, the pricing matrix and the mode picker are the two places that answer it, and they do not always answer it at the same time.

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 -- SpaceXAI's own. It is also, now, free. The upgrade you were told to wait for already happened, quietly, and the only thing you had to do to get it was nothing.

Corrections and Updates

October 2, 2026: Grok 4.6 is no longer the model grok.com names behind the paid modes. This page said grok.com's picker "names Grok 4.6 behind the paid Expert and Heavy modes." Read October 2, 2026 it names Grok 4.7 there, which SpaceXAI announced on September 21, 2026. The plan-inclusion claim is unchanged and re-verified: the pricing page's newest readable capture, September 28, 2026, still shows one model row reading Grok 4.6, ticked for all seven plans including Free, and the decode is sound because the Video generation row carries a cross in the Free column on the same read.

August 28, 2026: Grok 4.6 is on every consumer Grok plan, including Free, and this page said the opposite for thirteen days.