Skip to content
sobolewski.proAI today, every eveningSubscribe

Once a day. Unsubscribe with one click. Privacy policy

Claude API request size limit jumps from 32 MB to 128 MB

The Claude API request size limit is now 128 MB, four times more than before. That means more PDFs and images in one call. Here is where the new limits apply.

4 min read

This article in other languages:العربيةDeutschEspañolFrançaisItaliano日本語Polski简体中文

In this article
  1. What does the 128 MB request size limit change?
  2. Where do the new limits apply?
  3. Is this the end of Files API workarounds?
  4. What to watch out for
  5. FAQ

Anthropic has raised the Claude API request size limit for the Messages API and the Token Counting API from 32 MB to 128 MB. The change applies to the Claude API run directly by Anthropic and landed in the Claude Release Notes in the October 28 entry.

What does the 128 MB request size limit change?

In practice, it’s about how much data fits into a single POST /v1/messages call before the API answers with a 413 request_too_large error. The limit counts the whole request: text, conversation history, and images and PDFs sent as base64.

According to the vision docs, a single image can be up to 10 MB (after base64 encoding), and one API request can hold up to 600 images (100 for models with a 200K-token context window). Under the old limit, a few large images could eat up the 32 MB before you got anywhere near the image count limit. Now there’s four times as much headroom.

Same story with PDFs. According to the PDF support page, one request covers up to 600 pages (100 when the request’s context window is under 1M tokens), and both limits, pages and size, apply to the whole request. Long contracts or scans full of graphics, like the ones Claude for Legal is aimed at, are now easier to send in one piece.

Where do the new limits apply?

Only on the Claude API run by Anthropic (the so-called “first-party” API). The current table from the API overview looks like this:

Endpoint Maximum request size
Messages, Token Counting 128 MB
Message Batches API 256 MB
Files API 500 MB
Sessions, Agents, Environments 32 MB

Other platforms have different limits:

  1. Amazon Bedrock: 20 MB.
  2. Claude Models on Google Cloud (managed by Google): 30 MB.
  3. Claude Platform on AWS, Claude Models on Google Cloud (managed by Anthropic) and Microsoft Foundry: Messages and Token Counting still capped at 32 MB.

A fun detail from the API errors page: on the Claude API, the 413 error comes from Cloudflare before the request even reaches the API servers.

Is this the end of Files API workarounds?

Not necessarily. The docs still recommend sending large PDFs and many images through the Files API and referencing them by file_id. A single file there can be 500 MB, and the whole organization gets 1 TB of storage.

There are also a few constraints the bigger limit doesn’t change:

  1. When a request has more than 20 images, each of them is subject to a stricter dimension limit. The docs suggest keeping the long side under 3000 px or sticking to 20 image and document blocks.
  2. Dense PDFs can fill up the context window before you hit the page limit, even with the Files API.
  3. Agents in Claude Managed Agents (Sessions, Agents, Environments) stay at 32 MB.

(Request size isn’t the same as cost: every image and every PDF page turns into tokens, so it’s worth checking how much you actually use, like in my Claude Code token usage measurements.)

What to watch out for

First, the platform. If you use Claude through Bedrock, Google Cloud or Foundry, nothing has changed, and code that works on the Claude API may get a 413 there.

Second, token counting. The 128 MB limit also covers the Token Counting API, so you can count a big request before you send it.

Third, the missing details. The Release Notes entry is a single sentence and doesn’t say why the limit went up now, or whether partner platforms will see a similar change.

FAQ

What is the maximum request size in the Claude API?

On the Claude API run by Anthropic: 128 MB for Messages and Token Counting, 256 MB for Message Batches and 500 MB for the Files API. Sessions, Agents and Environments have 32 MB.

What does the 413 request_too_large error mean in the Claude API?

That the request exceeded the byte limit for that endpoint. The fix is to shrink the images, split the document, or upload the files through the Files API and reference them by file_id.

Does the 128 MB limit work on Amazon Bedrock?

No. According to the docs, Bedrock has a 20 MB limit, and Claude Platform on AWS has 32 MB for Messages and Token Counting.

How many PDF pages can I send to Claude in one request?

Up to 600 pages, or up to 100 when the request’s context window is under 1M tokens. The whole request also has to fit within the size limit.

  • AI
  • Anthropic
  • Claude

AI today, every evening

The day's biggest AI news and new posts from the blog, every evening. Short, with sources.

Once a day. Unsubscribe with one click. Privacy policy

4 min read

Claude Opus 4.8 fast mode ends on November 30

Claude Opus 4.8 fast mode goes away on November 30, and requests with speed: "fast" will return an error. Opus 5.5 offers it for less, with a few changes.

  • AI
  • Claude
  • Anthropic