Anthropic Lifts Claude API Requests to 128 MB, 6.4x Bedrock

Anthropic’s Claude API release notes, in an entry dated 28 October 2026, say: “We’ve raised the maximum request size for the Messages API and the Token Counting API from 32 MB to 128 MB on the first-party Claude API”. Anthropic’s developer account, @claudedevs, announced the change on X on 29 October, adding that “Now you get ~4x more room”.

The same documentation lists lower limits on the cloud platforms that resell Claude. By our arithmetic, the new first-party limit is 6.4 times the 20 MB Amazon Bedrock allows and four times the 32 MB on Microsoft Foundry.

Short explainer · Claude API requests rise to 128 MB

A vertical news Short narrated from this piece’s headline and opening line.

Business Pill · The Model and the Harness

A short explainer on the difference between an AI model and the harness around it. It teaches the general idea only and says nothing about the change in this piece.

What Changed

In its post, @claudedevs gave the reason: “Image-heavy requests like long computer-use sessions and big PDFs were previously hitting 32 MB with plenty of context window left”. In other words, by Anthropic’s account, the size of the request in megabytes, not the model’s context window, was the limit some customers hit first.

Anthropic’s API overview now lists 128 MB for Messages and Token Counting, 256 MB for the Message Batches API, 500 MB for the Files API and 32 MB for Sessions, Agents and Environments on the first-party Claude API. Requests above the limits receive “a 413 request_too_large error”. The Sessions, Agents and Environments endpoints still list 32 MB.

The Gap With the Cloud Platforms

The overview says partner-operated platforms set their own limits: “Bedrock limits requests to 20 MB, and Claude Models on Google Cloud (Google managed) limits requests to 30 MB”. It adds that “Claude Platform on AWS, Claude Models on Google Cloud (Anthropic managed), and Microsoft Foundry support Messages and Token Counting requests of up to 32 MB”, and tells developers to “Consult your platform’s documentation for current values”.

By our arithmetic, 128 MB is 6.4 times Bedrock’s 20 MB, about 4.3 times the 30 MB on Google-managed Google Cloud, and four times the 32 MB on Claude Platform on AWS, Anthropic-managed Google Cloud and Microsoft Foundry. Anthropic’s vision documentation also lists a smaller per-image ceiling on Bedrock: “10 MB (base64-encoded) on the Claude API” against “5 MB (base64-encoded) on Amazon Bedrock”.

Developers also reach Claude through those clouds. We reported that Amazon’s Bedrock tokens rose nearly 17x and covered the agent platforms at Amazon, Microsoft and Google in their October results.

What 128 MB Holds

Anthropic’s vision documentation puts the maximum size per image at 10 MB, base64-encoded, on the Claude API, and allows “600 for each request on the API, for all other models” other than those with a 200,000-token context window, which are allowed 100. By our arithmetic, at that 10 MB maximum a 32 MB request had room for three such images and a 128 MB request has room for twelve, before any other content. Smaller images fit in far greater numbers, up to the per-request count.

For documents, Anthropic’s PDF page lists a maximum of “600 (100 when the request’s context window is under 1M tokens)” pages per request and says that “Both limits are on the entire request payload, including any other content sent alongside PDFs”. It points to the Files API for large PDFs, “to keep request payloads small”.

Computer-use sessions are the other case Anthropic named. The vision documentation notes that, when a request holds more than 20 images, the images that count toward that threshold include “including images from earlier conversation turns that you resend and images nested inside `tool_result` content (for example, screenshots returned to the computer use tool)”. As we read it, a long agent session that resends its screenshots grows in megabytes with every step, which is the case @claudedevs named.

Connecting the Coverage

The change lands in a week of agent-tool news. Google’s exit guide for its Jules coding agent named Claude Code as its example, as covered in our piece on Google shutting Jules, and Microsoft’s Copilot offers Claude models alongside its own agent, as covered in our piece on Copilot and Claude Code.

The Structural Read

As we read Anthropic’s documentation, the first-party Claude API and the cloud platforms that resell Claude now accept different maximum request sizes, from 20 MB on Amazon Bedrock to 128 MB on Anthropic’s own API.

By our arithmetic, at the documented 10 MB maximum per image, a 32 MB request had room for three such images and a 128 MB request has room for twelve, before any other content.

Anthropic’s @claudedevs account named long computer-use sessions and big PDFs as the requests that hit the old 32 MB limit with context window left over.

Three Implications

FOR DEVELOPERS BUILDING AGENTS Anthropic’s documentation lists 128 MB for Messages requests on its own API, while its Sessions, Agents and Environments endpoints still list 32 MB.

FOR THE CLOUD PLATFORMS Anthropic’s documentation lists 20 MB on Amazon Bedrock, 30 MB on Google-managed Google Cloud and 32 MB on Microsoft Foundry, and tells developers to check each platform’s documentation for current values.

FOR LARGE DOCUMENTS Anthropic’s PDF documentation says the limits apply to the entire request payload and points to the Files API, which its overview lists at 500 MB, for large PDFs.

The Business Engineer Lens

This story maps onto the Business Engineer framework The Agentic Harness War.

The analysis puts it this way: “The agentic surface — the harness — is becoming where value pools in the entire AI stack.”

As we read it, the request size an API accepts is part of that surface: it sets how long an agent session can run before it has to be cut down. The analysis does not discuss this October 2026 change.

What Is Not Established

Anthropic did not say in the post or the release notes whether the partner platforms will raise their limits, how many customers hit the 32 MB ceiling, or whether pricing changes; tokens, not megabytes, set the price of a request. The comparison between platforms uses the values in Anthropic’s own documentation, which tells developers to check each platform’s documentation for current values. The image counts are our arithmetic at the documented maximum size, not a measured workload. We did not contact Anthropic, Amazon, Google or Microsoft.

The Bottom Line

Anthropic’s release notes say it raised the first-party Claude API’s maximum request size for Messages and Token Counting from 32 MB to 128 MB on 28 October, which @claudedevs called “~4x more room”. By our arithmetic, that is 6.4 times the 20 MB limit Anthropic’s documentation lists for Amazon Bedrock and four times the 32 MB listed for Microsoft Foundry.