Context window checker
Will this document fit in a model's context, with room for the answer?
FreeRuns in your browserNo signupUpdated 2026-10-29
Rates last checked October 2026. Window sizes change between model versions; check the provider's model page.
GPT-6.1 Sol
Fits
2,848 of 1,050,000 tokens used (0.3%)
Tokens to spare
1,047,152
Document 48 + system 800 + answer 2,000
| Model | Window | Used | Left | |
|---|---|---|---|---|
| GPT-6 Astra | 1,050,000 | 2,848 (0%) | 1,047,152 | Yes |
| GPT-6.1 Sol | 1,050,000 | 2,848 (0%) | 1,047,152 | Yes |
| GPT-6 Luna | 1,050,000 | 2,848 (0%) | 1,047,152 | Yes |
| Claude Opus 5.5 | 1,000,000 | 2,866 (0%) | 997,134 | Yes |
| Claude Sonnet 5.5 | 1,000,000 | 2,866 (0%) | 997,134 | Yes |
| Claude Haiku 5.5 | 1,000,000 | 2,866 (0%) | 997,134 | Yes |
| Claude Haiku 4.5 | 200,000 | 2,866 (1%) | 197,134 | Yes |
| Gemini 4 Argon | 1,000,000 | 2,848 (0%) | 997,152 | Yes |
| Gemini 3.8 Flash | 1,048,576 | 2,848 (0%) | 1,045,728 | Yes |
OpenAI and Gemini rows are counted with o200k_base, the published encoding for GPT-4o through GPT-5; GPT-6 and Gemini tokenizers differ slightly, so treat those rows as close estimates. Claude rows are an estimate at 3.5 characters per token. Quality often drops well before the window is full, so keep large requests under about 60% of the window where you can.
About this tool
How it works and how to read the result.
Need this done for you?
AI features for your product
Add AI search, chat and generation to your existing app.
See the serviceWhat it does
A context window calculator answers one question before you send a long document to a model: will it fit? Paste the text, choose a model, add the tokens your system prompt uses and the room you want for the answer. The page counts the document, adds the other two, and compares the total with each model's window. You see a fit, tight or too-big verdict for every model at once. Nothing leaves your browser.
How to use it
- Paste the document, transcript or code you want to send.
- Enter the token count of your system prompt. The token counter gives it to you.
- Enter the reserve for the answer: how long a reply you expect, in tokens.
- Pick your model. The table shows the same check for all listed models.
- If it does not fit, split the document or summarise sections before sending.
How to read the result
Say you paste a large set of documents that counts as 860,000 tokens, with a 4,000-token system prompt and 16,000 tokens reserved for the answer. That is 880,000 tokens. On GPT-6.1 Sol, with a 1,050,000-token window, it fits but uses 84%, so the row shows Tight. Gemini 4 Argon, at 1,000,000 tokens, is at 88%. On Claude Opus 5.5 the character-based estimate is higher, around 1,080,000 for the documents alone, so the row shows No. Claude Haiku 4.5, at 200,000 tokens, is far over.
Tight is a warning, not a failure. It means a slightly longer document or a longer conversation will break the request, and that accuracy may already be suffering. For a one-off summary it is fine; for a chatbot that accumulates history, leave much more room.
Tokens over tells you how much to cut. If the document is 20,000 tokens over, remove or summarise about 20% of it, or move to a model with a larger window.
Window sizes are listed on each provider's model page: OpenAI, Anthropic, Google. We handle long-document processing as part of AI product features.
Questions
Frequently asked.
What is a context window?
The maximum number of tokens a model can hold in one request, counting the system prompt, the conversation so far, any documents you attach and the reply it writes. Go over it and the request is rejected or older content is dropped. Most current models offer about a million tokens; a few smaller ones, such as Claude Haiku 4.5, offer 200,000.
Should I fill the window completely?
No. Models answer less accurately when the request is very long, and details in the middle of a huge prompt are easiest to miss. Keeping large requests under about 60% of the window leaves room for the reply and tends to give better answers.
Why do the counts differ between models?
Each provider has its own tokenizer. OpenAI and Gemini rows here use OpenAI's o200k_base encoding, which is close for GPT-6 and Gemini but not identical. Claude rows use a characters-divided-by-3.5 estimate because Anthropic's tokenizer is not available in the browser. Treat the Claude figure as approximate.
More in AI & automation
Related tools.
Your next move
Like what you see? Tell us about it.
A process that takes too long, a website that undersells you, or an idea ready to take shape. Tell us where you are and we will suggest a useful next step.
India · Working with clients worldwide