Usable context
Test the real LLM context window
Gradually increase context size and see how much this endpoint actually retains.
How this check works
An LLM context window test finds how much context an API relay actually keeps. The advertised window can be larger than the usable window after truncation or silent drops.
01Probe
Plant canary markers in a growing context payload.
02Inspect
Increase size until the endpoint drops or forgets a marker.
03Conclude
Report the last size that still retained the markers.
- How to read the result
- The usable window can be smaller than the advertised limit. A short window is a product risk for RAG and long agent traces, not just a number.
- When to run this check
- Run this before you buy a 128K or 200K route. A speed test will not catch silent truncation.
Recent reports
Public reports that included this check.
- sub.callai.onegpt-5.6-solInconclusiveReport timeAug 18
- sub.callai.onegpt-5.6-solBlockedReport timeAug 18
- ai.databyte.co.iddatabyte-m1BlockedReport timeAug 18
- ai.databyte.co.iddeepseek-v4-flashNormalReport timeAug 18
- ai.databyte.co.idMiniMax-M3NormalReport timeAug 18
- tokengate-cqt9ivzs.manus.spaceclaude-opus-5NormalReport timeAug 17
FAQ
What is an LLM context window test?
It measures the usable context the API relay actually keeps, not only the number shown on a marketing page.
Why can the real window be smaller?
A relay can truncate, summarize, or drop older tokens to save cost or stay inside an upstream limit.
Does this test prove the official model limit?
No. It proves the usable window on this endpoint at check time. Official model limits can still be different.