Grok Models Explained: Grok 4.6, 4.5, 4.3 & Grok 4 Heavy (2026)

xAI's Grok is a family of models, not one: Grok 4.6 (released August 12, 2026) is the newest flagship, an Opus-class model tuned for coding and agentic work, with a 500K context window; Grok 4.5 sits just below it (also 500K), Grok 4.3 holds 1M tokens, and Grok 4.20 and Grok 4.1 Fast reach 2M. Strikingly, the newest models have the smallest context of the recent lineup, so the right pick depends on your task and budget, not just recency. This guide breaks down the 2026 Grok lineup, context windows, pricing, and benchmarks, accurate as of August 2026 per xAI. AI Toolbox adds search, folders, and export on top of any Grok model.

By Adi Leviim, Co-Founder at Infi Developments, 7+ years building AI productivity tools.

What Grok Models Are Available in 2026?

As of August 2026, xAI keeps several Grok versions live at once with different context windows and price points. Grok 4.6 is the newest release, but the largest-context and cheapest options sit on other version numbers, which is why picking a model is not simply "use the newest."

• Grok 4.6 is the newest model, released August 12, 2026. It is xAI's current flagship, an Opus-class model tuned for coding and agentic work, with a 500K token context window and top-tier quality on xAI's benchmarks. On the API it is the top-quality choice at $2.00 / $6.00 per 1M tokens ($0.50 cached input).
• Grok 4.5, released July 8, 2026, was the prior flagship. Built on xAI's V9 architecture as a 1.5-trillion-parameter mixture-of-experts, it is an "Opus-class" model that is faster and cheaper, tuned for coding and agentic work and trained on real Cursor developer sessions, with the same 500K token context window. Independent testing by Artificial Analysis ranked it #4 of 168 models on its Intelligence Index (score 54) and first on agentic tool use.
• Grok 4.3, released April 2026 with a December 2025 knowledge cutoff, has a 1M token context window, native video input, document generation for PDFs, spreadsheets, and slides, improved tool-calling, and xAI's claim of a low hallucination rate.
• Grok 4.20 offers a larger 2M token context window and agentic variants with reasoning, non-reasoning, and multi-agent modes for long tasks.
• Grok 4.1 Fast is the cheap, high-throughput option, also with a 2M token context window, tuned for low-latency responses.
• Grok 4 Heavy is the multi-agent heavy tier for the most demanding problems, offered through xAI's top SuperGrok Heavy plan.
• Grok 5 has been reported in training on the Colossus 2 supercomputer, with an expected arrival later in 2026.

For the authoritative, current list and specs, xAI publishes its models at docs.x.ai. Whichever model you use, AI Toolbox for Grok works on top of it.

Grok Model Comparison (2026)

Prices reflect xAI's published rates as of August 2026 and can change; check docs.x.ai for the latest. xAI does not publish a separate maximum-output cap, so Grok output is bounded by the model's context window.

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

The move from Grok 4.3 to the 4.5 and 4.6 generation traded context length for reasoning quality and price efficiency. Grok 4.3 holds 1M tokens. Both newer flagships hold 500K, exactly half.

Grok 4.5 introduced xAI's V9 architecture as a 1.5-trillion-parameter mixture of experts. Because only a fraction of those parameters activate per token, it runs faster and cheaper than its size suggests. xAI trained it on real Cursor development sessions, which is why its strongest results are agentic and coding benchmarks rather than general knowledge.

Grok 4.6 kept that architecture and window while raising output quality. It is the model to reach for when the work is coding, tool use, or multi-step agent runs.

Grok 4.3 remains the better choice for two jobs the newer pair does not do as well. It accepts native video input, and it generates documents directly as PDFs, spreadsheets, and slides. If either matters, the newer flagship is a downgrade despite the higher version number. There is a fuller breakdown in our Grok 4.5 vs Grok 4.3 comparison.

The practical read: version number tracks reasoning quality, not capability breadth. Grok is one of the few model families where upgrading can cost you a feature.

Which Grok Model Should You Use?

Match the model to the job rather than always reaching for the flagship.

• Coding and agentic work: Grok 4.6, the newest flagship, is the pick, top-quality on agentic tool use, with Grok 4.5 a close, slightly cheaper alternative.
• Everyday reasoning and multimodal work with more context: Grok 4.3 gives you a 1M window with video input and document generation.
• Very long documents or agent runs: Grok 4.20 or Grok 4.1 Fast, because their 2M window is four times Grok 4.5's 500K.
• High volume on a budget: Grok 4.1 Fast is the cheap, fast option.
• The hardest, multi-step problems: Grok 4 Heavy on the SuperGrok Heavy plan.

On the consumer side, SuperGrok (around $30/month) provides access to the newest models, including Grok 4.6, with DeepSearch and Big Brain mode, while SuperGrok Heavy (around $300/month) adds Grok 4 Heavy. Pricing and access change often, so confirm on xAI's site.

How to Check Which Grok Model You Are Using

Grok shows the active model in the composer's model selector, on both x.com/i/grok and grok.com. Open the selector above the message box and the currently selected model is the one marked as active.

Two things catch people out. First, the two Grok surfaces are separate products sharing one account, and they do not always default to the same model. Our guide on Grok on X vs grok.com covers where they diverge.

Second, the model you pick is not the only thing shaping the answer. Your plan sets which models appear at all, and your surface sets how much context the model actually receives. A 2M-token model on the free web tier still works inside roughly 128K.

Through the API, the model name is returned on every response, so there is no ambiguity. xAI lists the current identifiers at docs.x.ai.

Context Windows: Why Bigger Is Not Always the Newest

A model's context window sets how much conversation and document text it can consider at once, and in Grok's lineup the newest models are the smallest. Grok 4.6 and Grok 4.5 hold 500K tokens, Grok 4.3 holds 1M, and Grok 4.20 and Grok 4.1 Fast hold 2M. The 4.5 and 4.6 generation halved the window from Grok 4.3's 1M in exchange for token efficiency, so if you routinely paste large codebases, long PDFs, or entire research folders into a single chat, an older 1M or 2M model can matter more than the newer, cheaper reasoning of Grok 4.6.

One important caveat: these are the models' maximum windows, available through the xAI API. The everyday Grok web interface on x.com/i/grok and grok.com enforces a smaller working window, around 128K tokens on the free tier, with larger windows on paid SuperGrok tiers. In other words, having a 2M-capable model does not mean the free web UI gives you 2M, so plan around the window your plan and surface actually provide.

No matter the window size, long chats still fill up, and Grok gives no native indication of how full a conversation is. The AI Toolbox context window meter shows tokens used and messages remaining so you can start a fresh chat before older context is dropped.

Which Grok Models Do Free and SuperGrok Plans Include?

Your xAI plan decides which models you can select, and the free tier is the most restricted on both model choice and context.

• Free Grok gives access to current general models with a working window of roughly 128K tokens and usage caps that reset on a rolling basis.
• SuperGrok (around $30/month) unlocks the newest flagships including Grok 4.6, plus DeepSearch and Big Brain mode, with a larger working window than the free tier.
• SuperGrok Heavy (around $300/month) adds Grok 4 Heavy, the multi-agent tier reserved for the hardest problems.

xAI changes plan contents and limits often, so treat those as a snapshot and confirm on xAI's own pricing page. We track the detail in our Grok pricing and API costs guide.

One limit no plan removes is history management. Grok has no folders, no full-text search across message bodies, and no export. That gap is identical on free and on SuperGrok Heavy, which is why a Grok Chrome extension is worth adding regardless of what you pay xAI.

Using AI Toolbox with Any Grok Model

AI Toolbox (formerly ChatGPT Toolbox) operates at the browser level on x.com/i/grok and grok.com, independent of which Grok model you select. It adds the productivity layer Grok lacks: full-text search across every message, folders and Smart Tags, Context Mentions (@@), a prompt library, and bulk export as TXT, Markdown, JSON, or PDF. Whether you run Grok 4.3 for reasoning or Grok 4.1 Fast for volume, your history stays searchable and organized. AI Toolbox is also available for ChatGPT, Gemini, and Claude.

Search is the feature that changes most as a Grok history grows. Grok's own history is a reverse-chronological list, so finding a specific answer from three weeks ago means scrolling. AI Toolbox indexes every synced message, so you can retrieve it by a phrase you remember. Our guide on finding a specific Grok message walks through the filters.

Organization works the same way on every model. Folders and subfolders group chats by project, and Smart Tags label them automatically by keyword. The free plan covers two folders; unlimited folders and bulk actions are Premium.

The rest of the layer is model-independent too: a prompt library with saved prompts, bulk export and delete in TXT, Markdown, JSON, and PDF, and the context meter. For a full walkthrough, see the Grok power-user guide and managing Grok chat history.

If you are still weighing Grok against other assistants, our Grok vs ChatGPT comparison covers where each one leads. AI Toolbox is one install that covers all four.

Whichever Grok model you use, don't lose your chats. AI Toolbox adds full-text search, folders, and export to Grok on X and grok.com. Trusted by 40,000+ users with a 4.6/5 Chrome Web Store rating. Compare Grok plans →

Frequently Asked Questions

What is the latest Grok model in 2026?

Grok 4.6, released August 12, 2026, is xAI's newest flagship. It is an Opus-class model tuned for coding and agentic work with a 500K token context window. Grok 4.5 from July 2026 sits just below it, and Grok 5 has been reported in training for later in 2026.

Which Grok model has the biggest context window?

Grok 4.20 and Grok 4.1 Fast both reach 2M tokens, larger than Grok 4.3's 1M and larger than Grok 4.6 and 4.5 at 500K. Those are API maximums. The Grok web interface works within a smaller window, roughly 128K tokens on the free tier.

How much does the Grok API cost?

As of August 2026, Grok 4.6 and 4.5 cost $2.00 per million input tokens and $6.00 output, with cached input at $0.50. Grok 4.3 and 4.20 run about $1.25 and $2.50, Grok 4.1 Fast about $0.20 and $0.50. Check docs.x.ai for current rates.

Is Grok 4.6 better than Grok 4.3?

For coding, tool use, and agentic runs, yes. For long documents, native video input, or generating PDFs and spreadsheets, Grok 4.3 is still the stronger choice. Grok 4.3 also holds 1M tokens against 500K, so upgrading the version number can cost you both context and features.

What is Grok 4 Heavy?

Grok 4 Heavy is xAI's multi-agent tier for the hardest problems, available only on the SuperGrok Heavy plan at around $300 per month. It runs several agents in parallel on one task rather than answering from a single pass, which suits complex multi-step research and engineering work.

Does Grok have folders or search for past chats?

No. Grok lists conversations in reverse chronological order with no folders, no full-text search inside message bodies, and no export. That gap is the same on the free tier and on SuperGrok Heavy, so it is a product limitation rather than a plan limitation you can pay to remove.

Does AI Toolbox work with every Grok model?

Yes. AI Toolbox runs at the browser level on x.com/i/grok and grok.com, so it is independent of the model you select. It adds full-text search, folders, Smart Tags, a prompt library, a context meter, and bulk export in TXT, Markdown, JSON, and PDF on top of any model.

Want to get more out of whichever Grok model you use? Install AI Toolbox free, open Grok, and search, organize, and export your chats. See AI Toolbox pricing and more Grok guides.

Last updated: August 28, 2026.