The limit is measured in bytes
Anthropic has raised the raw request-size ceiling for its Messages API and Token Counting endpoint from 32 MB to 128 MB, Emergent reports. The change addresses a transport constraint encountered by requests containing substantial image or document data.
This is separate from a model’s context window. A request can remain within its token allowance while exceeding the size allowed for the HTTP body. In that case, the reported failure is a 413 request-too-large error, rather than evidence that the model has exhausted its reasoning context.
Images make the distinction practical
The article highlights base64 encoding, which increases the size of binary attachments by roughly a third. Screenshots collected across a computer-use conversation can therefore consume the request budget even when the visible text is relatively short. A larger envelope gives those workflows more space before the same byte limit is reached.
That does not guarantee four times as many useful agent actions or successful document analyses. The number of screenshots, their sizes, the rest of the conversation and independent token constraints still matter. The update changes the permitted payload, not the correctness of the resulting answers.
Other endpoints retain separate ceilings
The report distinguishes the affected endpoints from Message Batches and the Files API, which have different limits. Teams should identify the exact route producing an error instead of treating every upload or batch operation as if it uses the Messages ceiling.
For a practical check, record the serialized request size alongside token usage when investigating failures. That helps separate a transport rejection from a model limit or application problem. A larger accepted request is useful operational headroom, but it does not by itself establish a pricing change, unlimited attachments or reliable completion of every larger task.