Guide · Reference

Context Window Comparison: 51 AI Models Ranked

The context window is how much text a model can consider at once — your prompt, documents and the reply all share it. Here is every supported model, largest first, with what fits inside.

Model Provider Context ≈ Words ≈ Pages
Gemini 1.5 Pro Google 2M tokens 1,500,000 3,000
DeepSeek V4 Flash DeepSeek 1.3M tokens 983,040 1,966
GPT-5.5 OpenAI 1.1M tokens 787,500 1,575
GPT-5.6 Sol OpenAI 1.1M tokens 787,500 1,575
GPT-5.6 Terra OpenAI 1.1M tokens 787,500 1,575
GPT-5.6 Luna OpenAI 1.1M tokens 787,500 1,575
Gemini 2.5 Flash Google 1.0M tokens 786,432 1,573
Gemini 3.5 Flash Google 1.0M tokens 786,432 1,573
Gemini 3.1 Pro Google 1.0M tokens 786,432 1,573
Gemini 3.6 Flash Google 1.0M tokens 786,432 1,573
Gemini 3.7 Flash Google 1.0M tokens 786,432 1,573
DeepSeek V4 Pro DeepSeek 1.0M tokens 786,432 1,573
MiniMax M3 MiniMax 1.0M tokens 786,432 1,573
Kimi K3 Moonshot / Kimi 1.0M tokens 786,432 1,573
GPT-4.1 OpenAI 1.0M tokens 785,682 1,571
GPT-4.1 mini OpenAI 1.0M tokens 785,682 1,571
Claude Sonnet 4.6 Anthropic 1M tokens 750,000 1,500
Claude Sonnet 5 Anthropic 1M tokens 750,000 1,500
Claude Opus 5 Anthropic 1M tokens 750,000 1,500
Claude Fable 5 Anthropic 1M tokens 750,000 1,500
Gemini 1.5 Flash Google 1M tokens 750,000 1,500
Gemini 2.0 Flash Google 1M tokens 750,000 1,500
Gemini Ultra Google 1M tokens 750,000 1,500
Grok 4.3 xAI / Grok 1M tokens 750,000 1,500
Grok 4.6 xAI / Grok 500K tokens 375,000 750
Mistral Small Mistral 262K tokens 196,608 393
Mistral Large 3 Mistral 262K tokens 196,608 393
Mistral Small 4 Mistral 262K tokens 196,608 393
Kimi K2.6 Moonshot / Kimi 262K tokens 196,608 393
Kimi K2.5 Moonshot / Kimi 262K tokens 196,608 393
Grok Build xAI / Grok 256K tokens 192,000 384
MiniMax M2.7 MiniMax 205K tokens 153,600 307
MiniMax M2 MiniMax 205K tokens 153,600 307
o1 OpenAI 200K tokens 150,000 300
o3 OpenAI 200K tokens 150,000 300
o3-mini OpenAI 200K tokens 150,000 300
Claude 3.5 Sonnet Anthropic 200K tokens 150,000 300
Claude 3.5 Haiku Anthropic 200K tokens 150,000 300
Claude 3 Opus Anthropic 200K tokens 150,000 300
Claude 3 Sonnet Anthropic 200K tokens 150,000 300
Claude Haiku 4.5 Anthropic 200K tokens 150,000 300
Llama 3.1 70B Meta / Llama 131K tokens 98,304 197
Llama 3.1 8B Meta / Llama 131K tokens 98,304 197
GPT-4o OpenAI 128K tokens 96,000 192
GPT-4o mini OpenAI 128K tokens 96,000 192
GPT-4 Turbo OpenAI 128K tokens 96,000 192
o1-mini OpenAI 128K tokens 96,000 192
Mistral Large Mistral 128K tokens 96,000 192
Llama 3.1 405B Meta / Llama 128K tokens 96,000 192
Mixtral 8x7B Mistral 32K tokens 24,000 48
GPT-3.5 Turbo OpenAI 16K tokens 12,289 25

What actually fits in a context window

Three practical notes

First, the window is shared: input + output must fit together, so a 128K model with a 30K prompt leaves ~98K minus any reserved output cap. Second, some providers apply tiered pricing above a threshold — the same token can cost more as your conversation grows. Third, bigger windows invite bigger prompts; pair them with prompt caching so repeated prefixes stay cheap, and estimate real workloads in the calculator's Chat tab.

Context window FAQ

Which AI model has the largest context window?

Gemini-class models lead with context windows of 1 million tokens and beyond — enough for roughly 750,000 words or 1,500 pages in a single request. Most frontier models from OpenAI and Anthropic sit between 128K and 200K tokens.

How many words is a 128K context window?

About 96,000 English words (~4 characters per token, 0.75 words per token) — roughly 190 pages of prose, a small codebase, or about 480 chat messages of 200 words each.

Does a bigger context window cost more?

Usually yes. Some providers charge more per token above certain context thresholds, and long prompts always cost proportionally more because you pay per token regardless of window size. Filling a 1M-token window is far pricier than filling 8K.

Do models actually use their full context well?

Not always. Effective recall degrades in the middle of very long contexts ("lost in the middle" effect). Best practice: keep critical instructions near the beginning or end of the prompt, even when the window is huge.