Context Windows
Data verifiedContext windows range from 32K to 1M tokens across today's AI models. The right size depends on your task — here's exactly what each model supports and when the larger context actually matters.
| Model | Provider | Context | ~Pages | Tier | Input $/M |
|---|---|---|---|---|---|
| Gemini 3.1 Pro | 2M | ~6,000 | Premium | $2.00 | |
| Grok 4 | xAI | 2M | ~6,000 | Balanced | $2.00 |
| GPT-5.6 Sol | OpenAI | 1M | ~3,150 | Premium | $5.00 |
| GPT-5.6 Terra | OpenAI | 1M | ~3,150 | Balanced | $2.00 |
| GPT-5.6 Luna | OpenAI | 1M | ~3,150 | Budget | $0.20 |
| Gemini 3.5 Flash | 1M | ~3,146 | Balanced | $1.50 | |
| Gemini 3.6 Flash | 1M | ~3,146 | Balanced | $1.50 | |
| Gemini 3.5 Flash-Lite | 1M | ~3,146 | Budget | $0.30 | |
| Muse Spark | Meta | 1M | ~3,146 | Balanced | $1.25 |
| Claude Opus 5 | Anthropic | 1M | ~3,000 | Premium | $5.00 |
| Claude Sonnet 5 | Anthropic | 1M | ~3,000 | Premium | $3.00 |
| DeepSeek V4-Pro | DeepSeek | 1M | ~3,000 | Budget | $0.43 |
| DeepSeek V4-Flash | DeepSeek | 1M | ~3,000 | Budget | $0.14 |
| Qwen 3.8 Max | Alibaba | 1M | ~3,000 | Balanced | $2.00 |
| Qwen 3.7 Max | Alibaba | 1M | ~3,000 | Balanced | $2.50 |
| Kimi K3 | Moonshot | 1M | ~3,000 | Premium | $3.00 |
| GLM-5.2 | Z.ai | 1M | ~3,000 | Budget | $1.40 |
| Claude Fable 5 | Anthropic | 1M | ~3,000 | Premium | $10.00 |
| Claude Mythos 5 | Anthropic | 1M | ~3,000 | Premium | $10.00 |
| Claude Opus 4.8 | Anthropic | 1M | ~3,000 | Premium | $5.00 |
| Claude Opus 4.7 | Anthropic | 1M | ~3,000 | Premium | $5.00 |
| GPT-5.5 | OpenAI | 1M | ~3,000 | Premium | $5.00 |
| Claude Opus 4.6 | Anthropic | 1M | ~3,000 | Premium | $15.00 |
| Claude Sonnet 4.6 | Anthropic | 1M | ~3,000 | Premium | $3.00 |
| Gemini 3.1 Flash | 1M | ~3,000 | Budget | $0.50 | |
| Llama 4 Scout | Meta | 512K | ~1,536 | Budget | $0.50 |
| Grok 4.5 | xAI | 500K | ~1,500 | Balanced | $2.00 |
| GPT-5.4 | OpenAI | 272K | ~816 | Premium | $2.50 |
| Mistral Medium 3.5 | Mistral | 256K | ~768 | Balanced | $1.50 |
| Kimi K2.7 Code | Moonshot | 256K | ~768 | Budget | $0.95 |
| Llama 4 Maverick | Meta | 256K | ~768 | Budget | $0.60 |
| Codestral 25.01 | Mistral | 256K | ~768 | Budget | $0.90 |
| GPT-5.2 | OpenAI | 200K | ~600 | Premium | $12.00 |
| Claude 4 Haiku | Anthropic | 200K | ~600 | Budget | $0.80 |
| GPT-5.2 Mini | OpenAI | 128K | ~384 | Balanced | $1.20 |
| GPT-4o | OpenAI | 128K | ~384 | Balanced | $5.00 |
| Mistral Large 2 | Mistral | 128K | ~384 | Balanced | $3.00 |
| DeepSeek V3 | DeepSeek | 128K | ~384 | Budget | $0.27 |
| DeepSeek R1 | DeepSeek | 128K | ~384 | Budget | $0.55 |
| GPT-4o Mini | OpenAI | 128K | ~384 | Budget | $0.15 |
| Mistral Small 3.1 | Mistral | 128K | ~384 | Budget | $0.10 |
Pages estimate assumes 250 words per page, 0.75 words per token.
Most tasks don't need 1M tokens. Here's a practical guide by task type.
Drafting a short email, answering a question, writing a code function. Any model works.
10–30 page reports, PR reviews, meeting transcripts, blog post drafts.
Book chapters, large technical specs, detailed research papers, mid-size codebases.
Refactoring full projects, analyzing complete repositories, multi-document synthesis.
A context window is the maximum amount of text an AI model can read and consider at once — your prompt, the model's previous responses, and any documents you include. It's measured in tokens (roughly 4 characters or 0.75 words each). A 1M token context window can hold approximately 750,000 words, or about 1,500 pages.
A short email is 100–300 tokens. A 10-page PDF is roughly 5,000–8,000 tokens. A full book chapter is 10,000–20,000 tokens. A complete novel is 100,000–200,000 tokens. A large codebase can be 200,000–800,000 tokens. For most tasks, even a 32K window is sufficient — 1M context matters for entire codebases or book-length research.
Gemini 2.0 Pro and Claude Sonnet 4.6 (Anthropic) both offer 1M token context windows — the longest currently available. This is enough to fit multiple books or an entire medium-sized codebase in a single prompt.
Not necessarily. A large context window allows more input, but models can struggle to use information from the very middle of very long contexts (the 'lost in the middle' problem). Claude Sonnet 4.6 is specifically noted for maintaining coherence across its full 1M context. Always test on your actual task.
You need a large context window when working with: entire codebases for refactoring or debugging, long legal or research documents, multi-turn conversations that must retain full history, or when combining multiple large files in one prompt. For single-file tasks or conversational use, 32K–128K is more than enough.
GPT-5.4 has a 272K token context window — large enough for most professional tasks, but about 3.7× smaller than Claude Sonnet 4.6 or Gemini 2.0 Pro's 1M token windows.