Which LLM has the largest context window?

Most current frontier models offer a 1M-token context window (1.0M), including the Claude, Gemini, GPT-4.1, DeepSeek, and Grok families. GPT-4o and Claude Haiku 4.5 are smaller (128K and 200K). Use the tool above to see exactly how your input fills each model's window.

What is a context window?

The context window is the maximum number of tokens a model can consider at once — it holds your system prompt, the conversation history, any documents or tool results, and the model's own response. Input and output share the same window, so a 1M-token window doesn't mean 1M tokens of input plus a full-length reply.

How many pages of text fit in a context window?

Roughly 650 tokens per page of prose (~500 words). So a 128K window holds about 200 pages, a 200K window about 300 pages, and a 1M window about 1,500 pages. Code is denser and fits fewer pages per token.

Does using the full context window cost more or slow the model down?

Yes. You pay input token rates for everything in the window, so a near-full 1M-token prompt is expensive, and larger prompts increase latency. Some providers also charge a premium tier for prompts above ~200K tokens. Keep only what the task needs in context.

Context Windows

LLM Context Window Comparison

Compare context windows across models and see how much of each window your input fills — with a fits / doesn't-fit flag.

Updated Jul 5, 2026FreeNo signup

Your input size (tokens)

Fits in 31 of 31 models

Gemini 3.1 Pro
1.0M
12%
Gemini 3.5 Flash
1.0M
12%
Gemini 3.1 Flash-Lite
1.0M
12%
Gemini 2.5 Pro
1.0M
12%
Gemini 2.5 Flash
1.0M
12%
Gemini 2.5 Flash-Lite
1.0M
12%
GPT-4.1
1.0M
12%
GPT-4.1 mini
1.0M
12%
GPT-4.1 nano
1.0M
12%
Claude Opus 4.8
1M
13%
Claude Opus 4.7
1M
13%
Claude Sonnet 5
1M
13%
Claude Sonnet 4.6
1M
13%
Claude Fable 5
1M
13%
DeepSeek V4 Flash
1M
13%
Grok 4.3
1M
13%
GPT-5.5
400K
32%
GPT-5.4
400K
32%
GPT-5
400K
32%
GPT-5 mini
400K
32%
GPT-5 nano
400K
32%
Codestral
256K
50%
o3
200K
64%
o4-mini
200K
64%
Claude Haiku 4.5
200K
64%
Llama 3.3 70B Instruct Turbo
131K
98%
Llama 3.1 405B Instruct Turbo
131K
98%
GPT-4o
128K
100%
GPT-4o mini
128K
100%
Mistral Large 3
128K
100%
Mistral Small 4
128K
100%

Bars show how much of each model's context window your 128,000-token input fills. Leave headroom for the model's response — the window is shared between input and output.

TL;DR

Most frontier models now share a 1M-token context window (Gemini 3.1 Pro, Gemini 3.5 Flash, Gemini 3.1 Flash-Lite and more), while GPT-4o sits at 128K and Claude Haiku 4.5 at 200K — GPT-4o is the smallest here at 128K. Enter a token count (or paste text) above to see how much of each model's window your input fills, with a fits / doesn't-fit flag. The full context-window table is below.

Context window by model

Updated Jul 5, 2026

Context window and max output tokens by model
Model	Provider	Context window	Max output
Gemini 3.1 Pro	Google	1,048,576(1.0M)	65,536
Gemini 3.5 Flash	Google	1,048,576(1.0M)	65,536
Gemini 3.1 Flash-Lite	Google	1,048,576(1.0M)	65,536
Gemini 2.5 Pro	Google	1,048,576(1.0M)	65,536
Gemini 2.5 Flash	Google	1,048,576(1.0M)	65,536
Gemini 2.5 Flash-Lite	Google	1,048,576(1.0M)	65,536
GPT-4.1	OpenAI	1,047,576(1.0M)	32,768
GPT-4.1 mini	OpenAI	1,047,576(1.0M)	32,768
GPT-4.1 nano	OpenAI	1,047,576(1.0M)	32,768
Claude Opus 4.8	Anthropic	1,000,000(1M)	128,000
Claude Opus 4.7	Anthropic	1,000,000(1M)	128,000
Claude Sonnet 5	Anthropic	1,000,000(1M)	128,000
Claude Sonnet 4.6	Anthropic	1,000,000(1M)	128,000
Claude Fable 5	Anthropic	1,000,000(1M)	128,000
DeepSeek V4 Flash	DeepSeek	1,000,000(1M)	384,000
Grok 4.3	xAI	1,000,000(1M)	—
GPT-5.5	OpenAI	400,000(400K)	128,000
GPT-5.4	OpenAI	400,000(400K)	128,000
GPT-5	OpenAI	400,000(400K)	128,000
GPT-5 mini	OpenAI	400,000(400K)	128,000
GPT-5 nano	OpenAI	400,000(400K)	128,000
Codestral	Mistral	256,000(256K)	—
o3	OpenAI	200,000(200K)	100,000
o4-mini	OpenAI	200,000(200K)	100,000
Claude Haiku 4.5	Anthropic	200,000(200K)	64,000
Llama 3.3 70B Instruct Turbo	Meta	131,072(131K)	—
Llama 3.1 405B Instruct Turbo	Meta	131,072(131K)	—
GPT-4o	OpenAI	128,000(128K)	16,384
GPT-4o mini	OpenAI	128,000(128K)	16,384
Mistral Large 3	Mistral	128,000(128K)	—
Mistral Small 4	Mistral	128,000(128K)	—

How it works

The visualizer computes fill% = input tokens ÷ context window for each model and flags whether your input fits. When you paste text, its tokens are counted in your browser with the GPT-4o tokenizer (a close proxy across models).

Context windows are the models' published limits, verified 2026-07-05. Remember the window is shared between input and output — leave room for the response, and note some providers meter prompts above ~200K tokens at a higher rate.

Sources

FAQ

Which LLM has the largest context window?: Most current frontier models offer a 1M-token context window (1.0M), including the Claude, Gemini, GPT-4.1, DeepSeek, and Grok families. GPT-4o and Claude Haiku 4.5 are smaller (128K and 200K). Use the tool above to see exactly how your input fills each model's window.
What is a context window?: The context window is the maximum number of tokens a model can consider at once — it holds your system prompt, the conversation history, any documents or tool results, and the model's own response. Input and output share the same window, so a 1M-token window doesn't mean 1M tokens of input plus a full-length reply.
How many pages of text fit in a context window?: Roughly 650 tokens per page of prose (~500 words). So a 128K window holds about 200 pages, a 200K window about 300 pages, and a 1M window about 1,500 pages. Code is denser and fits fewer pages per token.
Does using the full context window cost more or slow the model down?: Yes. You pay input token rates for everything in the window, so a near-full 1M-token prompt is expensive, and larger prompts increase latency. Some providers also charge a premium tier for prompts above ~200K tokens. Keep only what the task needs in context.