Anthropic raises Claude API request limit fourfold, from 32 MB to 128 MB

Anthropic has raised the maximum request size for the Claude API’s Messages and Token Counting endpoints from 32 MB to 128 MB, a roughly fourfold increase announced by the company’s ClaudeDevs account on X. Anthropic says the change addresses image-heavy requests, such as long computer-use sessions and big PDFs, that were hitting the 32 MB cap “with plenty of context window left”.

The change is narrow but practical. It does not alter what Claude models can read or reason over. It removes a transport limit that was cutting off requests well before the model reached its token budget.

What changed in the Claude API request limit

The announcement covers two endpoints on the first-party Claude API: Messages (POST /v1/messages), used to send prompts and conversations to Claude, and Token Counting (POST /v1/messages/count_tokens), used to count the tokens in a message before sending it. Both now accept requests up to 128 MB, according to Anthropic’s API overview page.

In a follow-up post, Anthropic added a second change: for requests with more than 20 images, the per-image limit rises from 2000 px to 3000 px. The same thread states that the 128 MB figure applies to the first-party Claude API, while access through cloud platforms is limited to between 20 and 32 MB.

Sources differ slightly on timing. Data Studios dates the request-size change to October 28, 2026, and says it appears in the Claude Platform release notes. The X post announcing it is dated October 29, 2026. Startup Fortune reported that Anthropic made the change through the X post rather than a blog announcement, and that the API documentation still showed the old 32 MB figure as of that week. The API overview page now lists 128 MB.

Why bytes, not tokens, were the bottleneck

The request size limit and the context window measure different things. The context window is the amount of text and images a model can reason over, counted in tokens. The request size cap counts the raw bytes of the HTTP payload sent to the API. Data Studios describes it as the maximum number of bytes a single request may contain, including the prompt text, conversation history and any base64-encoded images or PDFs embedded in the message.

Base64 is a way of encoding binary files, such as images and PDFs, as text so they can travel inside a JSON request. According to Startup Fortune, that encoding inflates file size by roughly a third, so a 24 MB PDF was already close to the 32 MB ceiling before any instructions or conversation history were added. Data Studios gives the same arithmetic: the old cap left room for about 24 MB of raw binary data in one request, while 128 MB allows roughly 96 MB. These are approximations before the prompt text is counted.

Computer-use agents hit the problem in a specific way. In these sessions, Claude takes repeated screenshots of a screen to decide its next action, and those screenshots accumulate inside a single request. Startup Fortune notes that this can happen even when the reasoning load barely registers against a 200,000-token or 1-million-token context window. The outlet also reports that developers had described the 32 MB cap in Anthropic’s public issue trackers as hard-stopping entire conversations rather than letting the model recover or truncate gracefully.

The full set of Claude API request limits

Only the Messages and Token Counting limits changed. The other first-party endpoint limits stay where they were, according to Anthropic’s documentation and Data Studios’ before-and-after comparison.

Endpoint (first-party Claude API)	Before	Now
Messages	32 MB	128 MB
Token Counting	32 MB	128 MB
Message Batches API	256 MB	256 MB
Files API	500 MB	500 MB
Sessions, Agents, Environments	32 MB	32 MB

A request that exceeds these limits returns a 413 request_too_large error. Data Studios reports that, according to Anthropic’s error documentation, Cloudflare returns this error on the first-party API before the request reaches Anthropic’s servers. An oversized request is therefore rejected at the network edge and does not consume model capacity.

Other multimodal limits still apply

Data Studios points out that other limits on multimodal requests did not change and will now be the binding constraint more often. Citing Anthropic’s documentation, it lists:

• A maximum of 10 MB per image, base64-encoded.
• A maximum of 8000 x 8000 pixels per image, or 3,000 pixels per side when a request contains more than 20 images.
• Up to 600 images per request for models with context windows above 200,000 tokens, and 100 for models with a 200k-token window.
• Up to 600 PDF pages per request, reduced to 100 pages when the request’s context window is under 1M tokens.

By the outlet’s count, at the maximum per-image size the old 32 MB limit fit three such images in a request, and the new limit fits twelve.

Where the 128 MB limit applies, and where it does not

Claude is offered through Anthropic’s own API and through cloud platforms. Anthropic-operated options include Claude Platform on AWS, Claude Models on Google Cloud (Anthropic managed) and Microsoft Foundry. Partner-operated options include Amazon Bedrock and Claude Models on Google Cloud (Google managed). Anthropic notes that feature availability varies by platform.

The higher cap applies only to the first-party API. Anthropic puts the cloud platform range at 20 to 32 MB. Data Studios, citing the API overview, gives the breakdown: Amazon Bedrock limits requests to 20 MB, Claude Models on Google Cloud (Google managed) to 30 MB, and the three Anthropic-operated cloud offerings support Messages and Token Counting requests of up to 32 MB.

For teams that deploy the same application across several environments, this widens the gap between them. Data Studios notes that code relying on the new ceiling will not port unchanged to a cloud marketplace deployment.

What the larger cap means for computer-use and document agents

Anthropic’s own summary is that developers now get “~4x more room”. Startup Fortune translates that into roughly four times as many screenshot-heavy turns in a computer-use session, or a meaningfully larger PDF attached to a single call, before the same wall returns. For document analysis tools and autonomous agents, the outlet expects fewer retries and fewer workarounds, such as splitting a PDF into separate requests only to stay under the byte limit.

Several limits are worth keeping in mind:

• Cost is unchanged. Coverage reports no change to model prices or rate limits, and no announced change to the context window. Data Studios cites Anthropic’s estimate that each PDF page typically uses 1,500 to 3,000 text tokens depending on density, with image token costs added because each page is also converted into an image. Larger requests remove a transport bottleneck, but the token bill still grows with page count and density.
• The Files API is still the recommended route for large inputs. According to Data Studios, the documentation continues to recommend uploading a PDF or image once and referencing it by file_id. This keeps payloads small, avoids re-sending the same bytes on every turn, and works within the separate 500 MB Files API limit. The outlet sees the higher inline ceiling as most useful for one-off jobs, stateless pipelines that cannot manage file lifecycles, and workloads that combine many medium-sized images or scanned pages in one call.
• Agent APIs keep the old cap. The beta Sessions, Agents and Environments APIs, which run Claude Managed Agents in cloud sandboxes, remain at 32 MB.

Developer reaction and open questions

On daily.dev, a community summary of about 200 comments describes developers as broadly welcoming the change as a fix for a real pain point. It also notes that many quickly raised caveats around billing, cloud-platform limits and base64 overhead. The site’s sentiment breakdown shows 40% positive, 45% mixed and 15% skeptical.

Those caveats match the open questions in the coverage. The sources do not say whether the cloud platforms will raise their own caps to match. Startup Fortune noted that Anthropic’s documentation briefly lagged the announcement. The update also arrived alongside a separate change: on October 29, Anthropic launched a Usage and Cost Analytics API in beta, which reports token usage and cost broken down by model, workspace, API key and the user or service account behind each request. For teams that send more data per call, that reporting may become more relevant.

What to watch next

• Whether Amazon Bedrock, Google Cloud and Microsoft Foundry raise their 20 to 32 MB limits toward the first-party 128 MB.
• Whether the Sessions, Agents and Environments APIs, still at 32 MB, receive a similar increase as Claude Managed Agents move out of beta.
• How often per-image, per-request image and PDF page limits now become the constraint that agent builders hit first.
• Whether the change reduces the developer complaints about hard-stopped conversations that Startup Fortune says had appeared in Anthropic’s issue trackers.