Grok 4 context window: 256,000 tokens

Grok 4, from xAI, has a context window of 256,000 tokens. The context window caps the total input the model can attend to in a single request. Anything past the limit has to be cut, summarized, or split across requests before the model sees it.

Verified 2026-08-10.

ModelContext window (tokens)
Grok 4256,000

Where 256K sits among current models

Grok 4 lands in the middle band of the current lineup. It is double the 128,000-token window of GPT-4o, close to Kimi K2.6 at 262,144, and above Claude Haiku 4.5 at 200,000. Within xAI's own lineup it is now the smallest window: Grok 4.5 takes 500,000 tokens and Grok 4.3 reaches 1,000,000, per xAI's model documentation.

Above it, GPT-5 offers 400,000 tokens, and the top of the table is the million-token group: GPT-5.5, the current Claude models, DeepSeek V4, and Grok 4.3 at 1,000,000, Gemini 2.5 Pro, Gemini 2.5 Flash, and Kimi K3 at 1,048,576, and GPT-5.6 Sol at 1,050,000. Grok 4 holds about a quarter of what those models accept in one request.

ModelContext window (tokens)
GPT-4o128,000
Claude Haiku 4.5200,000
Grok 4256,000
Kimi K2.6262,144
GPT-5400,000
Grok 4.5500,000
GPT-5.51,000,000
Claude Sonnet 51,000,000
DeepSeek V4 Flash1,000,000
Grok 4.31,000,000
Kimi K31,048,576
Gemini 2.5 Pro1,048,576
GPT-5.6 Sol1,050,000

Is 256K enough?

For most single-document work, yes. A 256K window handles long reports, sizable transcripts, and multi-file code questions in one request. The workloads that need more are the whole-corpus jobs: full repositories, large document collections, or long multi-session agent histories. Those are the cases where the million-token models earn their place.

Two general points apply at any size. Input billing scales with the tokens you send, so a fuller window is a costlier request. And long prompts can degrade retrieval of facts placed in the middle of the window, an effect well documented in long-context evaluations, so test at your real prompt length rather than assuming full-window accuracy.

For every current model side by side, see the context window comparison.