Under the hoodTier: how much it should matter to you
1Useful: my rating, 1 to 5
2Signal: worth watching, 1 to 5
ContextArea: what it touches
Internal ChangesKind: in v2.1.287,
Internal ChangesSection of the release
What
Before asking the API how many tokens (the units the model measures text in) some content uses, Claude Code makes its own rough estimate. It now compares that estimate against a second, higher limit. If the content is over that higher limit, the check answers yes straight away and the request to the API is never sent. Some parts of the content are left out of the estimate.
Before, only a lower limit was checked locally before the API request.
Why
Very large inputs no longer wait on a round trip to the API just to confirm they are too big.
How sure we are
Something disagreesSomething we can check disagrees with this entry, or the writer said they could not settle it.
The writer flagged doubtIt is not clear what Claude Code uses this token-count check for.
