AI & LLM
Token, cost & context tools for LLMs
Context Window Checker
Check each model’s context window and max output, see whether your prompt fits (with output budget reserved), and compare models side by side. Accurate, instant and free — for United States.
Tokenized in your browser — nothing is sent to a server. Leave blank to explore the window and headroom without a specific prompt.
GPT-5.5: 1,050,000-token context, 128,000 max output. Reserving 4,000 for the reply → 1,046,000 usable for input. Your ~0-token prompt fits (headroom 1,046,000). Exact count computed in your browser.
GPT-5.5 tokens
0
no prompt pasted
Headroom
1,046,000
tokens of headroom remaining
Input fit gauge
0 / 1,046,000 usable tokens
Fits — 0.000% of usable input used; 1,046,000 tokens of headroom remaining.
Usable input = 1,050,000 context − 4,000 reserved output = 1,046,000. Note: context window ≠ max output (128,000 for GPT-5.5) — these are separate caps.
All models — context window vs max output
Sorted by context window (largest first). Context window and max output are separate caps — a 1 M-context model can still cap its reply at 128K tokens.
| Model | Provider | Context window | Max output |
|---|---|---|---|
| GPT-5.5(selected) | OpenAI | 1,050,000 | 128,000 |
| Gemini 2.5 Pro | 1,048,576 | 65,536 | |
| Gemini 2.5 Flash | 1,048,576 | 65,536 | |
| Claude Opus 4.8 | Anthropic | 1,000,000 | 128,000 |
| Claude Sonnet 4.6 | Anthropic | 1,000,000 | 128,000 |
| GPT-5.4 mini | OpenAI | 400,000 | 128,000 |
| GPT-5.4 nano | OpenAI | 400,000 | 128,000 |
| Claude Haiku 4.5 | Anthropic | 200,000 | 64,000 |
| GPT-4o | OpenAI | 128,000 | 16,384 |
| GPT-3.5 Turbo | OpenAI | 16,385 | 4,096 |
| Text Embedding 3 Small | OpenAI | 8,192 | 8,192 |
Need an exact token count first? Use the Token Counter. Want to estimate API spend? Try the LLM API Cost Calculator.
How the fit check works
A model’s context window is the total token budget for one call — input and output share it. So checking fit isn’t just "is my prompt smaller than the window"; it’s "does my prompt fit once I’ve set aside room for the answer." This tool reserves an output budget, then tests the prompt against what remains.
Usable budget
window − output reserve
usable = window − output_reserve
Fits if prompt ≤ usable. Headroom = usable − prompt.
Two limits at once
window ≠ max output
Window = total in + out (e.g. 1M).
Max output = separate per-response cap (often 8K–128K). Both apply.
- 1Reserve output first, then check the prompt: usable = 1,000,000 − 32,000 = 968,000. The 800,000-token prompt is ≤ 968,000 → fits, with 168,000 tokens of headroom. Skip the reserve and you’d wrongly conclude a 990K prompt "fits" — leaving no room to answer.
- 2Pair it with count and cost: Get the prompt’s token count from the LLM Token Counter, then price the call with the LLM API Cost Calculator — fitting the window and affording it are separate questions.
Context window is not max output
What to do when the prompt overflows
If the prompt exceeds the usable budget, the call is rejected or truncated. The fixes: shorten or chunk the input, summarise earlier conversation turns, use retrieval to send only the relevant passages, or switch to a larger-window model. The checker shows exactly how many tokens you’re over.
A bigger window still costs per token
Frequently asked questions
The context window is the total token budget for one call — input and output combined. Reserve space for the answer, then check the prompt fits what’s left: usable = window − output_reserve, fits if prompt ≤ usable. An 800K prompt in a 1M window reserving 32K output → fits, 968K usable (168K headroom). Count your prompt with the LLM Token Counter.
No — and conflating them is the most common mistake. The context window is the total input + output budget (e.g. 1,000,000 tokens); the max output is a separate, much smaller cap on how many tokens the model can generate in one response (often 8,000–128,000). A model with a 1M-token window might still only emit 128K tokens per call. Both limits apply at once: your prompt must fit the window, and your requested output must fit the max-output cap.
Reserve at least the maximum length answer you expect, plus a small buffer. If you ask for a 4,000-token summary, reserve 4,000+; if you want the model to fill its whole output cap, reserve that cap. Whatever you reserve is subtracted from the window before checking the prompt fits, because input and output share the same budget — a prompt that fills the entire window leaves no room to answer.
The API rejects the call (or silently truncates, depending on the provider and settings) — you lose the overflowed tokens or get an error. The fix is to shorten or chunk the prompt, summarise earlier turns, use retrieval to send only relevant passages, or switch to a model with a larger window. The checker flags an over-limit prompt and shows how many tokens you need to cut.
Yes — fitting the window and paying for it are separate concerns. Every input token is billed at the model’s input rate regardless of headroom, so a prompt that fits comfortably can still be expensive. Use the context checker to confirm fit and the cost calculator to price it; a large window is an opportunity, not a free lunch.
Method, assumptions & references
Methodology: the context window is the combined input + output token budget; usable = window − output_reserve, and a prompt fits if prompt_tokens ≤ usable. Max output is a separate, smaller per-response cap — both limits apply. Worked example: an 800,000-token prompt reserving 32,000 output in a 1,000,000-token window is usable = 968,000 → fits with 168,000 headroom. Per-model window and max-output figures are dated and PR-updated on provider changes (repo-json-on-rate-change). Fitting the window is separate from the per-token cost of the call.
Cross-links
How we calculate this
Reviewed by Reckonist Editorial · Last reviewed 24 June 2026. Figures follow the methods and sources set out in our editorial standards.
Context window is the combined input + output token budget; max output is a separate, smaller per-response cap — both apply at once. Reserve output before checking a prompt fits. Per-model window and max-output figures are as of the review date and change with new model releases. Fitting the window is separate from the per-token cost of the call.
Keep going
Same-category tools follow this colour; a cross-category link keeps its own.