You'll noticeTier: how much it should matter to you
1Useful: my rating, 1 to 5
1Signal: worth watching, 1 to 5
ElsewhereArea: what it touches
Bug FixesKind: in v2.1.296,
Bug FixesSection of the release
What
Claude Code sometimes asks the API how many tokens a conversation uses. Tokens are the units the model reads text in. These requests include a setting for thinking, the model's step-by-step reasoning before it answers. That setting used to be fixed: thinking switched on, with a set token budget. It is now chosen for the model in use, so models that handle thinking differently get a setting that fits them.
Why
Token counts should be more accurate on newer models, and count requests are less likely to send a thinking setting that a model does not accept.
