/v1/messages - 413 Request too large - While under all limits.

[felix-locate]

413 "Request too large" error when referencing multiple files in Messages API despite being under documented limits

Description

I'm encountering a 413 Request too large error when using the /v1/messages API with multiple files uploaded via /v1/files API, despite being well under all documented limits.

Configuration

• Beta flags: context-1m-2025-08-07,files-api-2025-04-14
• Model: claude-sonnet-4-20250514
• SDK: @anthropic-ai/sdk v0.68.0

The Issue

When referencing 4 PDF files in a single message request, I receive a 413 error even though:

• ✅ Total PDF file size: 6.49 MB (well under any documented payload limit)
• ✅ HTTP request payload: ~18 KB (just JSON with file references)
• ✅ Token usage: ~271K tokens (well under 1M context window)
• ✅ Individual files: All under 100 MB per-file limit
• ✅ Each file uploads successfully to /v1/files

Reproduction Steps

1. Upload 4 PDF files via /v1/files API:

   • File 1: 1.76 MB (87 pages)
   • File 2: 3.44 MB (20 pages)
   • File 3: 951 KB (8 pages)
   • File 4: 364 KB (4 pages)
   • Total: 6.49 MB, 119 pages

2. Send a message referencing all 4 files:

{
  "model": "claude-sonnet-4-20250514",
  "max_tokens": 1024,
  "messages": [
    {
      "role": "user",
      "content": [
        {
          "type": "text",
          "text": "Summarize these files"
        },
        {
          "type": "document",
          "source": {
            "type": "file",
            "file_id": "file_011CVynXuw9EEpoFvNbA9La9"
          }
        },
        {
          "type": "document",
          "source": {
            "type": "file",
            "file_id": "file_011CVynXz3hWHxVDkiZSKJeT"
          }
        },
        {
          "type": "document",
          "source": {
            "type": "file",
            "file_id": "file_011CVyntPf4PXtRDZym7sqcD"
          }
        },
        {
          "type": "document",
          "source": {
            "type": "file",
            "file_id": "file_011CVyo1fyHG9DSBNDD5TgSL"
          }
        }
      ]
    }
  ]
}

3. Receive error:

{
  "type": "error",
  "error": {
    "type": "invalid_request_error",
    "message": "Request too large"
  },
  "request_id": null
}

Testing Results

Through systematic testing, I've identified an undocumented threshold:

Test	Files	Total Size	Status	Tokens Used
1	1 file	3.44 MB	✅ Success	238,708
2	2 files	5.2 MB	✅ Success	250,846
3	3 files	6.13 MB	✅ Success	270,752
4	4 files	6.49 MB	❌ 413 Error	N/A
5	3 files	8.65 MB	❌ 413 Error	N/A

Finding: There appears to be an undocumented limit of ~6.3 MB on the combined size of all files referenced in a single Messages API request.

Expected Behavior

Based on the documentation:

• Files API supports up to 100 MB per file
• Messages API supports up to 32 MB total request size
• With context-1m-2025-08-07 beta, I have a 1 million token context window

I expected my 6.49 MB request with 271K tokens to succeed.

Actual Behavior

Requests fail with 413 when the combined referenced file size exceeds ~6.3 MB, regardless of:

• The actual HTTP payload size (~18 KB)
• Token usage (well under 1M limit)
• Individual file sizes (all under 100 MB)

Environment

• Platform: Node.js
• SDK Version: @anthropic-ai/sdk@0.68.0
• Request method: REST API via proxy server

Additional Context

The error occurs during the file processing/text extraction phase on Anthropic's side, not during HTTP transmission. The 413 error suggests a resource constraint when processing multiple large PDFs simultaneously, separate from token limits.