Images can now be limited by token count, and one model's image limit rises to 2576px
A new maxImageTokens limit lets image resizing cap images by token count, and one model's limit moves from 2000x2000 to 2576x2576 with a 4784-token cap
You'll noticeTier: how much it should matter to you
3Useful: my rating, 1 to 5
3Signal: worth watching, 1 to 5
ImagesArea: what it touches
Internal ChangesKind: in v2.1.299,
What probably matters to youSection of the release
Unclear It is not clear where maxImageTokens is set or whether anything applies it yet.
What
When you give Claude Code an image, it may shrink the image before sending it to the model. Until now it only kept the width and height under a maximum. It can now also keep the image under a token budget. Tokens are the units the model counts its input in, so a bigger image costs more tokens.
A new resize helper first clamps the width and height to the model's limits, then, if maxImageTokens is set, shrinks the image further until it fits the token budget.
Resize reporting now records whether an image was resized for tokens, along with the target width and height and the limits that applied.
Why
For the model with the new limits, images can be sent at a higher resolution than before. At the same time, a large image may now be scaled down even when it is under the pixel limits, because it would cost too many tokens. If an image's fine detail matters, that budget is what decides how much of it reaches the model.
