Grok Models Explained (2026): Grok 4.7, 4.6, 4.5, 4.3, 4.20 and Build

xAI's newest Grok model is Grok 4.7, released September 21, 2026 with a 500K token context window, on the API, in Cursor and in Grok Build. The rest of the lineup is Grok 4.6 (August 2026, 500K), Grok 4.5 (July 2026, 500K), Grok 4.3 and Grok 4.20, which hold 1M tokens, and Grok Build 0.1, a 256K coding model. In the Grok apps, xAI's plan comparison still lists Grok 4.6 on every plan, including Free. Older models such as Grok 4, Grok 4.1 Fast and Grok 3 were retired from the API on May 15, 2026 (docs.x.ai and x.ai, October 2026). The newest model is not the one with the biggest context window, which is the most common mistake when choosing.

This guide explains which Grok models exist as of October 5, 2026, what each one is for, their context windows and reasoning settings, when each was released, which ones come with which Grok plan, and how to pick. It also covers xAI's image, video and voice models, and what happened to the names you may still see in older articles. Every specification comes from xAI's model pages, release notes and announcements, linked in the sources at the end. Prices are not repeated here; they are in our Grok pricing and plans guide.

By Adi Leviim, Co-Founder at Infi Developments. 7+ years building AI productivity tools. Specifications checked against docs.x.ai and x.ai on October 5, 2026.

Which Grok Models Are Available Today?

xAI's model documentation lists six Grok text models on the API: Grok 4.7, Grok 4.6, Grok 4.5, Grok 4.3, Grok 4.20 (in reasoning, non-reasoning and multi-agent versions) and Grok Build 0.1. All six accept text and image input and return text. xAI's own advice on its models page is short: "For everything else, including code, use Grok 4.7. It is the most capable model we've built."

xAI's release notes say both Grok 4.7 and Grok 4.6 have "no text output limit." Like every Grok model, they do not know current events unless you enable the Web Search or X Search tools. For per-model token prices, see the Grok API pricing table.

Grok Model Release Timeline

xAI released or retired a Grok model roughly every month in 2026, which is why so many articles still name models that no longer exist. These dates come from xAI's API release notes and news posts.

1. July 9, 2025: Grok 4 launches, alongside a new SuperGrok Heavy tier with Grok 4 Heavy.
2. November 19, 2025: Grok 4.1 Fast becomes available in the xAI Enterprise API.
3. March 10, 2026: Grok 4.20 and Grok 4.20 Multi-Agent go live on the API.
4. May 15, 2026: xAI retires grok-4-0709, the Grok 4.1 Fast and Grok 4 Fast models, grok-code-fast-1 and grok-3 from the API. Requests to those names are redirected to Grok 4.3, or to Grok Build 0.1 for grok-code-fast-1.
5. May 19, 2026: Grok Build 0.1, xAI's coding model, enters early access.
6. July 8, 2026: Grok 4.5 arrives on the API (xAI's announcement post is dated July 16).
7. July 22, 2026: Grok 4.5 rolls out on grok.com, X and the iOS and Android apps.
8. August 7, 2026: Imagine Image 2.0 is announced.
9. August 12, 2026: Grok 4.6 is released; it is the model xAI's plan comparison names on every Grok plan.
10. September 21, 2026: Grok 4.7 is released on the xAI API, in Cursor and in Grok Build, at the same price as Grok 4.6, with a Grok 4.7 Fast variant at twice the price in Cursor and Grok Build only.
11. November 2, 2026 (scheduled): grok-imagine-image-quality is retired in favor of grok-imagine-image-2.0.

Grok 4.7 vs Grok 4.6 vs Grok 4.5 vs Grok 4.3: What Changed

Grok 4.7, Grok 4.6 and Grok 4.5 trade half of Grok 4.3's context window for stronger coding and agent performance. Grok 4.3 holds 1M tokens; the three newer models hold 500K.

Grok 4.7 uses "a new, larger base model compared to Grok 4.6," with a longer reinforcement learning run weighted toward tasks that take many hours, and xAI says it is better at verifying its own work and managing longer context. It is served "at the same price and speed as Grok 4.6." In xAI's announcement table, Grok 4.7 at xhigh effort scores 46.3% on CursorBench 4.0 against 40.4% for Grok 4.6 at high, 37.6% on Terminal-Bench 4.0 against 20.3%, and 71.0% on DeepSWE v1.1 (at high effort) against 65.2%; on GDPval it reaches an Elo of 1695 against 1605. These are xAI's published figures, not independent tests.

Grok 4.5 was, in xAI's words, "trained alongside Cursor" on datasets spanning coding, science, engineering and math, with reinforcement learning across hundreds of thousands of multi-step software tasks. xAI says it is served at about 80 tokens per second and used about 4.2 times fewer output tokens than Opus 4.8 (max) per SWE Bench Pro task (15,954 against 67,020 on average). Its reported scores include 83.3% on Terminal Bench 2.1 and 64.7% on SWE Bench Pro.

Grok 4.6 builds on Grok 4.5 "with a particular focus on long-running agents and more ambitious interactive and visual work." xAI says it had a longer supplemental training run and adds more self-testing on long tasks. In xAI's own evaluation table, Grok 4.6 High scores 61 on the Artificial Analysis Intelligence Index against 56 for Grok 4.5 High, 69.9% on CursorBench 3.2 against 66.7%, and 65.9% on DeepSWE 1.1 against 54%. These are xAI's published figures, not independent tests.

Grok 4.3 remains useful for jobs the newer pair does not cover. It has the 1M window, a "none" reasoning setting for fast, cheap non-reasoning calls, and Batch API support; Grok 4.7, Grok 4.6 and Grok 4.5 do not accept batch requests. It is also where most retired model names now redirect. Our older Grok 4.5 vs Grok 4.3 comparison goes deeper on that pair.

Which Grok Model Should You Use?

Pick by the job and the size of your input, not by the version number.

• Software engineers and coding agents: Grok 4.7, xAI's recommended code model, through the API, Cursor or Grok Build. Grok Build 0.1 is the lighter coding option inside a 256K window.
• Analysts feeding in very large documents or codebases: Grok 4.3 or Grok 4.20, because a 1M window takes twice what Grok 4.7 or Grok 4.6 accepts in one request.
• Researchers running deep, multi-source investigations: Grok 4.20 Multi-Agent on the API, or SuperGrok Heavy in the Grok apps, where xAI describes a larger team of agents per answer.
• Developers processing large volumes overnight: Grok 4.3 or Grok 4.20 through the Batch API, the only Grok text models with a batch discount.
• Latency-sensitive apps that do not need reasoning: Grok 4.3 with reasoning set to none, or Grok 4.20 non-reasoning.
• Everyday users of grok.com or X: you do not choose an API model; xAI's plan comparison lists Grok 4.6 for the Grok apps on every plan.

Grok Context Windows: Why the Newest Model Is Not the Largest

A context window is how much text a model can consider in one request, and in Grok's lineup Grok 4.3 and Grok 4.20 hold 1M tokens, twice Grok 4.7, Grok 4.6 and Grok 4.5 at 500K, while Grok Build 0.1 holds 256K. If you routinely paste long contracts, research folders or whole repositories, a 1M model can matter more than the newest one.

Two details change the maths. First, xAI bills every token in a request at higher long context rates once a prompt reaches 200K tokens, on every Grok text model, so a large window is available but not cheap to fill. Second, the numbers above are API limits. xAI does not publish a token figure for the Grok apps; its plan page only describes relative lengths, with SuperGrok Lite offering "2x longer conversations in Chat" and SuperGrok "5x longer conversations in Chat." Earlier figures such as "2M tokens for Grok 4.20" or "128K on the free web app" do not appear on xAI's current model, pricing or plan pages.

Long Grok chats still fill up without warning. The AI Toolbox context window meter estimates how full a conversation is and how many messages remain, so you can move to a fresh chat before earlier context drops out. To measure a prompt before sending it, use the token counter.

Which Grok Models Come With Each Plan?

In the Grok apps, the model is the same on every plan: on October 5, 2026, xAI's plan comparison listed Grok 4.6 for Free, SuperGrok Lite, SuperGrok, SuperGrok Plus, SuperGrok Heavy, Business and Enterprise. What changes with the plan is how much you can use it and which modes and tools come with it.

• Free: Grok 4.6, image generation, Grok Build and limited real-time web and X search. No Expert mode, no video generation, no Grok Bot.
• SuperGrok Lite: adds Expert mode, which xAI describes as access to "our latest, best model," plus video generation.
• SuperGrok: Expert mode with higher limits, and Grok Bot agents.
• SuperGrok Plus: significantly higher usage, 1080p video and priority access.
• SuperGrok Heavy: "most powerful intelligence," with a larger team of agents collaborating to answer. xAI's current pages do not name a separate Grok 4 Heavy model for this tier.
• X Premium+: includes SuperGrok access on X, according to the X Help Center.

Monthly and annual prices for each tier, and how X Premium fits in, are in the Grok pricing guide.

Grok Imagine, Voice and Other xAI Models

Beyond the text models, xAI's docs list dedicated models for images, video and voice. xAI's recommendation table maps images to Grok Imagine Image 2.0, video to Grok Imagine Video 1.5 and voice to the Grok Voice API.

• Grok Imagine Image 2.0: image generation and editing with low, medium or auto quality at 1K or 2K, and up to five source images per edit since August 28, 2026.
• Grok Imagine Video 1.5: text-to-video, image-to-video and reference-to-video with optional preset voices, up to 1080p.
• Grok Voice Think Fast 2.0: xAI's speech-to-speech model, released July 29, 2026, with separate speech-to-text and text-to-speech APIs.

How Grok Model Names and Aliases Work

Every Grok API model has a fixed name plus aliases that move to newer versions automatically. xAI's convention is that the plain model name points to the latest stable version, a name ending in "-latest" points to the latest version, and a name ending in a date points to one specific release that never changes.

Retired names keep working but are served by a different model. Since May 15, 2026, a request to a retired reasoning model is answered by Grok 4.3 with low reasoning effort, and a request to a retired non-reasoning model by Grok 4.3 with no reasoning, billed at Grok 4.3 rates.

If your code still calls one of these names, set the replacement model explicitly so you control the reasoning effort you pay for, as xAI recommends in its migration guide.

Using AI Toolbox With Any Grok Model

AI Toolbox (formerly ChatGPT Toolbox) is a Chrome extension that works inside Grok on x.com/i/grok and grok.com, so it behaves the same whichever model or plan you use. It adds tools for managing your conversation history rather than changing the model: full-text search across your synced Grok messages, folders and subfolders, Smart Tags, a prompt library, the context meter, and export as TXT, Markdown, JSON or PDF.

Premium for the Grok module is $9.99 a month or $99 lifetime, and the purchase is linked to your X handle. All Access unlocks the ChatGPT, Gemini, Claude and Grok modules for $199 once; it is an extension license, not an AI subscription. For step-by-step guides, see finding a specific Grok message, managing Grok chat history and the Grok power-user guide. Still choosing an assistant? Our Grok vs ChatGPT and Grok vs Claude comparisons cover where each leads.

Limitations and What xAI Documents Well

xAI's developer documentation is thorough for API models: each model page states the context window, reasoning settings, batch support and rate limits. That makes the API side of this guide easy to verify. The gaps are elsewhere. xAI publishes no release date for Grok 4.3 in its release notes, no token figure for the context the Grok apps give each plan, and no independent benchmark data; the scores quoted above are xAI's own. None of the xAI pages cited here mention a Grok 5 release, so this guide does not predict one.

Model names change quickly at xAI, and aliases can silently point to a newer model, so check the models page at docs.x.ai before hard-coding a model into production. AI Toolbox does not change which model Grok uses or what your plan allows.

Frequently Asked Questions

What is the latest Grok model?

Grok 4.7 is the latest Grok model. xAI released it on September 21, 2026 as its frontier model for coding, agentic tasks and knowledge work, with a 500K token context window, on the xAI API, in Cursor and in Grok Build. In the Grok apps, xAI's plan comparison still lists Grok 4.6, released August 12, 2026, on every plan including Free.

Which Grok model has the biggest context window?

Grok 4.3 and Grok 4.20 have the biggest context windows among current Grok models, at 1M tokens each. Grok 4.7, Grok 4.6 and Grok 4.5 hold 500K tokens, and Grok Build 0.1 holds 256K. These are API limits from docs.x.ai; xAI does not publish a token figure for the Grok apps.

Is Grok 4.6 better than Grok 4.3?

For coding and agent tasks, Grok 4.6 is the stronger of the two in xAI's positioning, and Grok 4.7 has since replaced it as xAI's recommended model. Grok 4.3 still has advantages: a 1M token window against 500K, a no-reasoning mode, and Batch API support, which Grok 4.6 does not offer. For very large inputs or overnight batch jobs, Grok 4.3 can be the better fit.

What happened to Grok 4.1 Fast and Grok 4?

xAI retired Grok 4 (grok-4-0709), Grok 4.1 Fast, Grok 4 Fast and Grok 3 from its API on May 15, 2026. Requests to those model names still work but are served by Grok 4.3 and billed at Grok 4.3 rates. grok-code-fast-1 now redirects to Grok Build 0.1.

Is Grok 4 Heavy still available?

Grok 4 Heavy launched with the SuperGrok Heavy tier in July 2025, but xAI's current plan pages no longer name it. SuperGrok Heavy is now described as the most powerful tier, with a larger team of agents collaborating on each answer. On the API, the closest option is Grok 4.20 Multi-Agent.

Can free Grok users use Grok 4.6?

Yes. xAI's plan comparison lists Grok 4.6 on the Free plan as well as on every paid tier. Free users get lower usage and limited web and X search, and they do not get Expert mode, video generation or Grok Bot, which start on SuperGrok Lite or SuperGrok.

Does AI Toolbox work with every Grok model?

Yes. AI Toolbox runs in your browser on x.com/i/grok and grok.com, so it does not depend on the model or plan you use. It adds full-text search, folders, Smart Tags, a prompt library, a context meter and export on top of Grok: TXT of any single chat for free, and Markdown, JSON and PDF on Premium.

Related reading: Grok pricing and plans, Grok on X vs grok.com, AI model comparison table, and AI Toolbox for Grok.

Sources

• xAI Docs, Models (with the individual model pages for Grok 4.7, 4.6, 4.5, 4.3, 4.20 and Build 0.1)
• xAI Docs, API release notes
• xAI Docs, Model retirement on May 15, 2026
• xAI, Introducing Grok 4.7 (September 21, 2026)
• xAI, Introducing Grok 4.6 (August 12, 2026), plus xAI's Grok 4.5 announcements of July 16 and July 22, 2026
• xAI, Pricing and plan comparison, and the plan cards on grok.com/plans

Last updated: October 5, 2026