Anthropic Quadruples Claude API Request Size to 128 MB for Image-Heavy Workloads

Anthropic has increased the maximum request size for the Claude API’s Messages and Token Counting endpoints from 32 MB to 128 MB. The fourfold increase gives image-heavy conversations, large PDFs, and other multimodal workloads more room in a single request.

More bytes, same context window

The previous 32 MB ceiling could reject image-heavy requests even when their content fit within a model’s context window. Long computer-use sessions might include dozens of screenshots, while PDFs can carry high-resolution figures and scanned pages. Developers often had to resize images, split documents, trim conversation history, or upload content through the Files API.

The new limit applies to POST requests sent to /v1/messages and /v1/messages/count_tokens at api.anthropic.com , as described in Anthropic’s request-size documentation . Request size includes the serialized payload and encoding overhead, so base64-encoded media can consume substantially more space than the original files.

The model’s context window and token limits remain unchanged. A 128 MB payload can still exceed the selected model’s token capacity, and applications must continue to manage conversation history and document length accordingly.

Higher-resolution image batches

Anthropic has also raised the per-image resolution cap for requests containing more than 20 images, from 2,000 pixels to 3,000 pixels. The higher cap lets vision pipelines preserve more detail in screenshots, dense tables, technical diagrams, and scanned documents that previously required more aggressive resizing.

Cloud routes keep smaller limits

The 128 MB allowance applies only to Anthropic’s first-party API. Requests routed through Amazon Bedrock, Google Cloud, Microsoft Foundry, or other hosted Claude offerings remain subject to platform-specific limits.

Applications that support several providers should enforce the limit for the active route before sending a request. A payload accepted by api.anthropic.com may still fail on Bedrock or Google Cloud.

Workloads that gain headroom

The larger payload allowance is live on the first-party Claude API. It requires no new SDK version or beta header, and Anthropic has announced no related pricing change. Existing clients can use the additional capacity as long as they do not impose their own lower request-body limits.

• Computer-use agents can retain more screenshots in conversation history.
• Multimodal PDF workflows can process larger documents without splitting related pages.
• Vision pipelines can send more than 20 images at up to the higher 3,000-pixel cap.
• Token-counting requests can inspect larger multimodal payloads before inference.

Requests above a platform’s limit typically return HTTP 413 with a request_too_large error. Payloads exceeding 128 MB still need another path: the Message Batches API accepts requests up to 256 MB, while the Files API guide documents uploads up to 500 MB.