Context Window Checker
Paste text or drop a file and see which chat apps, API models and local models can take all of it.
No text yet, so this shows a 50-page document. Drag the slider, or paste text or drop a file for a real count.
Chat apps
- Claude Paid plans - Opus 5 / 5.5, Sonnet 5 / 5.5, Fable 5.1 1M≈4%Fits, using about 4% of 1,000,000 tokens
- Gemini Google AI Pro / Ultra 1M≈4%Fits, using about 4% of 1,000,000 tokens
- Claude Paid plans - Fable 5, Opus 4.6–4.8, Sonnet 4.6 500K≈7%Fits, using about 7% of 500,000 tokens
- ChatGPT Pro - Reasoning 400K≈9%Fits, using about 9% of 400,000 tokens
- ChatGPT Go / Plus - Reasoning 256K15%Fits, using about 15% of 256,000 tokens
- Claude Paid plans - other models 200K≈19%Fits, using about 19% of 200,000 tokens
- ChatGPT Pro - Instant 128K29%Fits, using about 29% of 128,000 tokens
- Gemini Google AI Plus 128K≈29%Fits, using about 29% of 128,000 tokens
- ChatGPT Go / Plus - Instant 54K69%Fits, using about 69% of 54,000 tokens
- Gemini No AI plan (free) 32K+5.3KToo long by about 5,333 tokens
- ChatGPT Free - Instant 27K+10KToo long by about 10,333 tokens
API
- GPT-6 Astra 1.05M≈4%Fits, using about 4% of 1,050,000 tokens
- GPT-6 Luna 1.05M≈4%Fits, using about 4% of 1,050,000 tokens
- GPT-6.1 Sol 1.05M≈4%Fits, using about 4% of 1,050,000 tokens
- Gemini 2.5 Flash 1.05M in≈3%Fits, using about 3% of 1,048,576 tokens
- Gemini 2.5 Flash-Lite 1.05M in≈3%Fits, using about 3% of 1,048,576 tokens
- Gemini 2.5 Pro 1.05M in≈3%Fits, using about 3% of 1,048,576 tokens
- Gemini 3 Flash Preview 1.05M in≈3%Fits, using about 3% of 1,048,576 tokens
- Gemini 3.1 Flash-Lite 1.05M in≈3%Fits, using about 3% of 1,048,576 tokens
- Gemini 3.1 Pro Preview 1.05M in≈3%Fits, using about 3% of 1,048,576 tokens
- Gemini 3.5 Flash 1.05M in≈3%Fits, using about 3% of 1,048,576 tokens
- Gemini 3.5 Flash-Lite 1.05M in≈3%Fits, using about 3% of 1,048,576 tokens
- Gemini 3.6 Flash 1.05M in≈3%Fits, using about 3% of 1,048,576 tokens
- Gemini 3.7 Flash 1.05M in≈3%Fits, using about 3% of 1,048,576 tokens
- Gemini 3.8 Flash 1.05M in≈3%Fits, using about 3% of 1,048,576 tokens
- Claude Fable 5.1 1M≈4%Fits, using about 4% of 1,000,000 tokens
- Claude Opus 5.5 1M≈4%Fits, using about 4% of 1,000,000 tokens
- Claude Sonnet 5.5 1M≈4%Fits, using about 4% of 1,000,000 tokens
- Claude Haiku 4.5 200K≈19%Fits, using about 19% of 200,000 tokens
Local
- DeepSeek-R1 0528 8B 131K≈28%Fits, using about 28% of 131,072 tokens
- DeepSeek-R1 Distill 1.5B–70B 131K≈28%Fits, using about 28% of 131,072 tokens
- Dolphin 3.0 8B 131K≈28%Fits, using about 28% of 131,072 tokens
- Mistral Nemo 12B 131K≈28%Fits, using about 28% of 131,072 tokens
- Phi-3 medium 14B 131K≈28%Fits, using about 28% of 131,072 tokens
- Phi-4 mini 3.8B 131K≈28%Fits, using about 28% of 131,072 tokens
- Qwen3 0.6B–32B 41K≈91%Fits, using about 91% of 40,960 tokens
- Mistral 7B 33K+4.6KToo long by about 4,565 tokens
- Mistral Small 22B–24B 33K+4.6KToo long by about 4,565 tokens
- Qwen2.5 0.5B–72B 33K+4.6KToo long by about 4,565 tokens
- Qwen2.5 Coder 1.5B–32B 33K+4.6KToo long by about 4,565 tokens
- Phi-4 14B 16K+21KToo long by about 20,949 tokens
- SmolLM2 360M–1.7B 8.2K+29KToo long by about 29,141 tokens
- OLMo 2 7B–13B 4.1K+33KToo long by about 33,237 tokens
- Phi-3 mini 3.8B 4.1K+33KToo long by about 33,237 tokens
- TinyLlama 1.1B 2K+35KToo long by about 35,285 tokens
| Model or plan | Type | Tokens | ≈ Pages | Source |
|---|---|---|---|---|
| GPT-6 Astra | API | 1,050,000 | 1,575 | Oct 8, 2026 |
| GPT-6 Luna | API | 1,050,000 | 1,575 | Oct 8, 2026 |
| GPT-6.1 Sol | API | 1,050,000 | 1,575 | Oct 8, 2026 |
| Gemini 2.5 Flash | API · input limit | 1,048,576 | 1,573 | Oct 8, 2026 |
| Gemini 2.5 Flash-Lite | API · input limit | 1,048,576 | 1,573 | Oct 8, 2026 |
| Gemini 2.5 Pro | API · input limit | 1,048,576 | 1,573 | Oct 8, 2026 |
| Gemini 3 Flash Preview | API · input limit | 1,048,576 | 1,573 | Oct 8, 2026 |
| Gemini 3.1 Flash-Lite | API · input limit | 1,048,576 | 1,573 | Oct 8, 2026 |
| Gemini 3.1 Pro Preview | API · input limit | 1,048,576 | 1,573 | Oct 8, 2026 |
| Gemini 3.5 Flash | API · input limit | 1,048,576 | 1,573 | Oct 8, 2026 |
| Gemini 3.5 Flash-Lite | API · input limit | 1,048,576 | 1,573 | Oct 8, 2026 |
| Gemini 3.6 Flash | API · input limit | 1,048,576 | 1,573 | Oct 8, 2026 |
| Gemini 3.7 Flash | API · input limit | 1,048,576 | 1,573 | Oct 8, 2026 |
| Gemini 3.8 Flash | API · input limit | 1,048,576 | 1,573 | Oct 8, 2026 |
| Claude Fable 5.1 | API | 1,000,000 | 1,500 | Oct 8, 2026 |
| Claude Opus 5.5 | API | 1,000,000 | 1,500 | Oct 8, 2026 |
| Claude Paid plans - Opus 5 / 5.5, Sonnet 5 / 5.5, Fable 5.1 | Chat apps | 1,000,000 | 1,500 | Oct 8, 2026 |
| Claude Sonnet 5.5 | API | 1,000,000 | 1,500 | Oct 8, 2026 |
| Gemini Google AI Pro / Ultra | Chat apps | 1,000,000 | 1,500 | Oct 8, 2026 |
| Claude Paid plans - Fable 5, Opus 4.6–4.8, Sonnet 4.6 | Chat apps | 500,000 | 750 | Oct 8, 2026 |
| ChatGPT Pro - Reasoning | Chat apps | 400,000 | 600 | Sep 26, 2026 |
| ChatGPT Go / Plus - Reasoning | Chat apps | 256,000 | 384 | Sep 26, 2026 |
| Claude Haiku 4.5 | API | 200,000 | 300 | Oct 8, 2026 |
| Claude Paid plans - other models | Chat apps | 200,000 | 300 | Oct 8, 2026 |
| DeepSeek-R1 0528 8B | Local | 131,072 | 197 | Oct 2, 2026 |
| DeepSeek-R1 Distill 1.5B–70B | Local | 131,072 | 197 | Oct 2, 2026 |
| Dolphin 3.0 8B | Local | 131,072 | 197 | Oct 2, 2026 |
| Mistral Nemo 12B | Local | 131,072 | 197 | Oct 2, 2026 |
| Phi-3 medium 14B | Local | 131,072 | 197 | Oct 2, 2026 |
| Phi-4 mini 3.8B | Local | 131,072 | 197 | Oct 2, 2026 |
| ChatGPT Pro - Instant | Chat apps | 128,000 | 192 | Sep 26, 2026 |
| Gemini Google AI Plus | Chat apps | 128,000 | 192 | Oct 8, 2026 |
| ChatGPT Go / Plus - Instant | Chat apps | 54,000 | 81 | Sep 26, 2026 |
| Qwen3 0.6B–32B | Local | 40,960 | 61 | Oct 2, 2026 |
| Mistral 7B | Local | 32,768 | 49 | Oct 2, 2026 |
| Mistral Small 22B–24B | Local | 32,768 | 49 | Oct 2, 2026 |
| Qwen2.5 0.5B–72B | Local | 32,768 | 49 | Oct 2, 2026 |
| Qwen2.5 Coder 1.5B–32B | Local | 32,768 | 49 | Oct 2, 2026 |
| Gemini No AI plan (free) | Chat apps | 32,000 | 48 | Oct 8, 2026 |
| ChatGPT Free - Instant | Chat apps | 27,000 | 41 | Sep 26, 2026 |
| Phi-4 14B | Local | 16,384 | 25 | Oct 2, 2026 |
| SmolLM2 360M–1.7B | Local | 8,192 | 12 | Oct 2, 2026 |
| OLMo 2 7B–13B | Local | 4,096 | 6.1 | Oct 2, 2026 |
| Phi-3 mini 3.8B | Local | 4,096 | 6.1 | Oct 2, 2026 |
| TinyLlama 1.1B | Local | 2,048 | 3.1 | Oct 2, 2026 |
Where the windows come from
- Chat apps: the context window each plan gets, from OpenAI’s, Anthropic’s and Google’s help pages. These are the same figures as our token counter and are re-checked every day.
- API models: the context window on each model’s own documentation page, re-read every day. Google lists the Gemini API’s input and output limits separately; those rows show the input limit and are marked “in”.
- Local models: the longest context the publisher’s config file gives, for the models in our “What AI can my computer run?” list. Sizes of the same model share a row.
How the check works
Your text is counted with OpenAI’s o200k_base tokenizer in your browser. Each bar
is a model’s whole window on one scale; the line is your text, and the coloured part is what
your text plus the room you leave for the reply would take. Every step on the scale is four
times the one before, so a 2,000-token prompt and a 1-million-token book fit on the same
chart. Pages assume 500 words a page and OpenAI’s rule of thumb of 100 tokens ≈ 75 words. See
our methodology.
Frequently asked questions
What is a context window?
The most text, measured in tokens, that an AI model can work with at once. It covers your messages, any files you add and, for most models, the reply as well. Text past the limit is cut off or the request is refused, and in a long chat the earliest messages drop out.
How many pages fit in a 1 million token context window?
Roughly 1,500 single-spaced pages of English. OpenAI’s rule of thumb is that 100 tokens is about 75 words, so 1 million tokens is about 750,000 words, and at 500 words a page that’s 1,500 pages. Google gives a similar figure for Gemini. Code, tables, numbers and most other languages use more tokens per word, so fewer pages fit.
Why does the reply count against the window?
For OpenAI and Anthropic models the published context window is a total for what you send and what the model writes back, so the room you leave for the reply comes out of it. Google publishes the Gemini API’s input limit and output limit separately, so for those rows the reply setting doesn’t change the bar.
Is the token count exact?
Only where a row is not marked “≈”. We count with OpenAI’s o200k_base tokenizer, which runs in your browser. Claude, Gemini and local models use their own tokenizers, and OpenAI doesn’t say which one its newest API models use, so for those the count is an estimate and the real figure can be noticeably higher or lower. If a result is close to the limit, leave some room.
Does my file get uploaded?
No. Text files are read by your browser and PDFs are opened with pdf.js on your own device. Nothing you paste or drop is sent to us or stored. Scanned PDFs without a text layer can’t be read, because there’s no text in them, only pictures of pages.
Why might a local model not take as much as shown?
The figure is the longest context the publisher’s own config file says the model supports. Ollama and LM Studio load a model with whatever context length the app is set to, and a longer context needs more memory. Our “What AI can my computer run?” tool shows how much memory each context length takes on your graphics card or Mac.
Did this tool do what you needed?
Related tools
Count words, characters and tokens, and see how much of each chatbot’s limit you use.
ChatGPT, Claude and Gemini paid plans side by side: price, limits and features.
Pick your graphics card or Mac and see which local AI models fit in its memory.