Context Window Checker
Check if a system prompt, retrieved chunks and the expected answer fit a model’s context window. Shows percent used, a warning colour and spare chunk capacity.
Private by design: everything you enter is processed in your browser. It is not sent to our servers.
How to use
- Pick a model. Its context window and maximum output size fill in automatically; you can overwrite the context limit to test a smaller budget.
- Enter the token size of your system prompt, how many retrieved chunks you add and how big each is, the user question, and the answer length you expect.
- Read the percentage used. Green is under 70%, amber 70–90%, red 90–100%, and “over” means the request will not fit.
- Use “most chunks that fit” to size your retrieval step, and copy a share link to keep the setup.
Worked example
A RAG pipeline sends a 1,000-token system prompt, eight 500-token chunks and a 200-token question, and allows 1,000 tokens for the answer. The model has an 8,000-token window (a small custom limit, to show how the warning colours behave).
| System prompt | 1,000 tokens |
|---|---|
| Chunks | 8 × 500 tokens |
| User question | 200 tokens |
| Expected output | 1,000 tokens |
| Context limit | 8,000 tokens |
Result
| Input total | 5,200 tokens |
|---|---|
| Input + output | 6,200 tokens |
| Used | 77.5% (amber) |
| Remaining | 1,800 tokens |
| Most chunks that fit | 11 (3 more than now) |
How it works
Total = system tokens + chunks × chunk tokens + user tokens + expected output tokens. The percentage is total ÷ context limit. Providers count the generated answer inside the same window, which is why output tokens are included.
The warning colours are this page’s own thresholds, not provider rules: under 70% is green, 70% to under 90% amber, 90% to 100% red, above 100% “over”. They are meant to leave room for token-count error and for longer-than-expected answers.
“Most chunks that fit” is floor((limit − system − user − output) ÷ chunk size). If a chunk size of zero is entered the result is unlimited.
Context limits and maximum output sizes are read from the data file, where each value records the provider page it came from and the date it was checked. If expected output exceeds the model’s maximum output, you get a separate warning.
Common errors
contextLimit must be at least 1.- Enter a positive context limit, or pick a model from the list.
chunks must be a whole number of 0 or more.- Use whole numbers only; remove decimals and negatives.
Over the context window by N tokens.- Remove chunks, shorten the system prompt, reduce expected output, or choose a model with a larger window.
Expected output (N) is above the model's maximum output (M).- The model cannot produce that many tokens in one response. Lower the expected output or split the task.
Limits
- You must supply token counts. This page does not tokenize text; use the Token Cost Calculator or RAG Chunker to measure it.
- It checks the total against the window only. Real requests also include formatting overhead, tool definitions and hidden system content that you should add to the system prompt figure.
- Some providers charge more or behave differently for very long prompts; a prompt that fits is not necessarily cheap.
- Only models with an official, dated context-window source are listed.
Privacy
Only numbers are processed, in your browser. Nothing is sent anywhere. No account is needed, no cookies are set and no analytics run. See thePrivacy Policy. The “Copy share link” button puts your settings after the “#” in the address; browsers do not send that part to servers, and the page’s canonical address never includes it.
FAQ
Does the output count against the context window?
Yes. The window covers the prompt and the generated answer together, so the calculator adds expected output to the input total.
Why include a safety margin?
Token counts are often estimates, and answers can run longer than planned. Staying under about 70% of the window leaves room for both.
What is “most chunks that fit”?
The largest number of equal-sized chunks that can be added without going over the window, after reserving space for the system prompt, question and output.
Can I test a window that is not listed?
Yes. Overwrite the context limit field with any number.
Related
Embed this tool
Paste this snippet into your own page or docs to show the tool inline. The embedded view works the same way (everything runs in the visitor’s browser) and carries a small link back to MakeTools.
Sources
- ai.google.dev/gemini-api/docs/models/gemini-3.5-flash-lite – checked 2026-10-14
- developers.openai.com/api/docs/models – checked 2026-10-14
- github.com/ggml-org/llama.cpp/blob/master/tools/quantize/README.md – checked 2026-10-14
- huggingface.co/Qwen/Qwen2.5-14B-Instruct/raw/main/config.json – checked 2026-10-14
- huggingface.co/Qwen/Qwen2.5-32B-Instruct/raw/main/config.json – checked 2026-10-14
- huggingface.co/Qwen/Qwen2.5-7B-Instruct/raw/main/config.json – checked 2026-10-14
- huggingface.co/api/models/Qwen/Qwen2.5-14B-Instruct – checked 2026-10-14
- huggingface.co/api/models/Qwen/Qwen2.5-32B-Instruct – checked 2026-10-14
- huggingface.co/api/models/Qwen/Qwen2.5-7B-Instruct – checked 2026-10-14
- huggingface.co/api/models/microsoft/Phi-3.5-mini-instruct – checked 2026-10-14
- huggingface.co/api/models/mistralai/Mistral-7B-Instruct-v0.3 – checked 2026-10-14
- huggingface.co/microsoft/Phi-3.5-mini-instruct/raw/main/config.json – checked 2026-10-14
- huggingface.co/mistralai/Mistral-7B-Instruct-v0.3/raw/main/config.json – checked 2026-10-14
- platform.claude.com/docs/en/about-claude/models/overview – checked 2026-10-14
Page last updated 2026-10-14. Written by Prajyot Mulik; see the editorial policy.