·5 min read
Claude API Raises Its Request Limit to 128 MB: PDFs, Images, and 413 Errors

When working with PDFs, multiple images, or browser screenshots, you may encounter a frustrating problem: Claude still has room in its context window, but your request is rejected because the payload is too large.
A recent Claude API update eases this restriction. Claude’s official developer account announced that the maximum request body size for Messages and Token Counting has increased from 32 MB to 128 MB, providing more room for image-heavy tasks and large PDF requests. Read the official announcement.
What does this change mean for document analysis and computer-use applications? Can you now submit a 128 MB PDF directly? Here is what developers need to know.
Which Claude API endpoints support the 128 MB limit?
According to the current documentation, the first-party Claude API has the following maximum request sizes:
- Messages: 128 MB
- Token Counting: 128 MB
- Message Batches: 256 MB
- Files: 500 MB
These limits apply to different endpoints and are not interchangeable. A larger upload limit on the Files API does not give a Messages request the same allowance. If you access Claude through a cloud platform or third-party gateway, check that platform’s own limits. See the Claude API documentation.
Request size vs. context window: what is the difference?
Request body size measures transmitted bytes. The context window measures the number of tokens a model can process.
Your application must satisfy both constraints: the request body must fit within the endpoint’s byte limit, and the input plus required output space must fit within the model’s context requirements.
The 128 MB update does not mean the context window has become four times larger. A request can fit within the byte limit yet exceed the available context. Conversely, it can have room left in the context window but exceed the payload limit because of embedded image data.
Track the serialized request size and estimated input tokens separately rather than checking only one of them.
Can you send a 128 MB PDF directly?
Not necessarily. If you embed a PDF in JSON as Base64, the transmitted payload includes the encoded file, prompts, message history, and other fields. Base64 typically increases binary data size by about one-third.
For example, a roughly 90 MB file could become approximately 120 MB after encoding, before other request content is added. This is an encoding-size estimate, not a measured upload result.
PDFs must also meet format, page-count, and context requirements. For large documents, Anthropic recommends considering the Files API and referencing the upload by file_id to keep the Messages payload small. File references still remain subject to model processing limits. Read the PDF support documentation.
Which applications benefit most?
Multiple-image and screenshot analysis
Tasks such as product-image comparisons and interface reviews may require multiple images in one request. A larger payload allowance can reduce the need to split requests solely because of image data size.
Keep only the images and resolution the task needs. Whether additional images improve the answer should be evaluated on your own workload.
Computer-use workflows
Browser and desktop automation generate screenshots over time. If your application retains many of them in the conversation, subsequent requests can keep growing.
The official announcement specifically mentions long computer-use sessions. The larger allowance provides more headroom, while applications still need to manage screenshot history and remove irrelevant material.
PDF document analysis
Documents containing scanned pages, charts, and illustrations can produce large data payloads. Developers can revisit earlier document-splitting strategies while continuing to check page-count and context requirements.
How to fix Claude API 413 request_too_large errors
A 413 request_too_large error means the request exceeds the permitted byte size. On the first-party Claude API, Cloudflare can return this error before the request reaches the API servers. See the official error documentation.
- Confirm your actual endpoint. Check whether a cloud platform, reverse proxy, or third-party gateway sits between your application and Claude. Other services in the chain may impose their own limits.
- Measure the final request body. Check the byte size of the serialized JSON, not just the original file size.
- Reduce embedded data. Remove duplicate screenshots and irrelevant message history. Compress images or split documents when appropriate.
- Consider file references. For suitable large documents, evaluate uploading through the Files API and referencing the upload.
- Recheck model constraints. After fixing the byte-size issue, confirm token, page-count, and other input requirements.
Repeatedly retrying the same oversized request usually will not resolve the problem. Adjust the payload or the relevant infrastructure configuration first.
Frequently asked questions
Does the 128 MB update improve answer quality?
It increases the amount of data a request can carry. Any improvement in answer quality depends on the relevance of that data and how your prompts and workflow are organized.
Will streaming fix a 413 error?
Streaming primarily changes how the response is returned. For a 413 caused by an oversized input payload, you still need to reduce or restructure the input.
If Token Counting accepts 128 MB, will Messages always succeed?
No. Token Counting helps estimate input tokens. A generation request must still meet payload, context, output configuration, and model-specific requirements.
Does every Claude provider automatically support 128 MB?
Check your provider’s documentation. The first-party Claude API’s limits do not automatically apply to all cloud platforms and gateways.
How should you update an existing application?
If your application previously enforced a fixed 32 MB cap, review that configuration and adjust it to match your actual API endpoint. Before deploying, validate request sizes, token usage, response times, and error handling with representative PDFs and images.
The most immediate benefit of the 128 MB update is fewer transport-size restrictions on large multimodal requests. Selecting relevant inputs, managing conversation history, and recording payload size remain essential to reliable Claude API integrations.
Based on official announcements and documentation checked on October 30, 2026. This article does not include independent performance benchmarks.
Find a model API that fits your AI application
Alongside request limits, API pricing and access options are worth evaluating. Tokatlas is a paid multi-model API platform: choose a model and set a target discount or price ceiling, and the platform looks for available API routes that meet your criteria. Matching depends on availability.
Visit Tokatlas to explore available models and pricing and find an API access option for your next AI project.