ModelsReference

Context windows

How much a model can read in one request, and how to check it.

By the end: You can size a request to a model's context window without guessing.

A context window is the total number of tokens a model can consider in one request: your input plus its output together. Exceed it and the request is rejected upstream.

Checking a model's window

The models page shows one number per model. On that page you can also run a context check, which asks the model itself to confirm its window — a single click, with no large prompt sent.

Practical sizing

Reserve room for the reply: a 1M window with a 990K prompt leaves little for output.
Token counts are approximate for non-English text and code; leave headroom.
Long documents plus long conversations overflow faster than either alone.

Counting before sending

The Anthropic-compatible counting route reports input tokens without generating anything:

Count tokens

curl https://www.ary4n73.com/v1/messages/count_tokens \
  -H "x-api-key: $ROUTEFORGE_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{"model":"claude-sonnet-4-6","messages":[{"role":"user","content":"Count these tokens"}]}'