Service eligibility and regional restrictions
PuppyIP serves only compliant overseas businesses and their authorized personnel. Proxy services are not available in mainland China. The service may only be used for lawful business activities outside mainland China. Use of this service within mainland China is prohibited.
Hosting a proxy IP or server overseas does not change these restrictions. The service must not be provided to end users in mainland China through relaying, forwarding, sharing or resale. Before use, read the Terms of Service.
Key Takeaways
- On October 28, 2026, Anthropic raised the first-party Claude API request-size limit for Messages and Token Counting from 32 MB to 128 MB.
- First-party Message Batches, Files, Sessions, Agents, and Environments endpoints have separate limits of 256 MB, 500 MB, and 32 MB.
- Bedrock, Google-managed Claude on Google Cloud, and Anthropic-managed cloud platforms have different request-size limits; the first-party 128 MB figure does not apply to all of them.
- HTTP 413 request_too_large is about request bytes. HTTP 429 rate_limit_error concerns rate or spend limits. The reviewed docs do not define an exact MB conversion or every client's serialization accounting.
Where does the 128 MB limit apply?
On October 28, 2026, Anthropic raised the first-party Claude API request-size limit for Messages and Token Counting from 32 MB to 128 MB. This changes request size only; the announcement does not raise context, output-token, or rate limits.
The announcement is limited to the first-party API. Claude on Bedrock, Google Cloud, or Anthropic-operated cloud offerings is a different platform with its own published limits.
Docs state MB but do not define the exact byte conversion or every SDK's serialized-body accounting. Do not infer a precise byte threshold. The stated maximum applies to its listed endpoint and platform, not every route or account that can serve Claude.
Check the endpoint and hosting platform
The first-party overview lists Messages and Token Counting at 128 MB; Message Batches at 256 MB; Files at 500 MB; and Sessions, Agents, and Environments at 32 MB. These are endpoint-specific caps.
Hosted limits differ: Amazon Bedrock is 20 MB; Google-managed Claude on Google Cloud is 30 MB. Messages and Token Counting support up to 32 MB on the Anthropic-operated Claude Platform on AWS, Anthropic-managed Claude on Google Cloud, and Microsoft Foundry.
The reviewed Google-managed product documentation calls it part of Google Cloud Agent Platform. This review did not verify account access, regional eligibility, or fees; confirm the exact offering and its current documentation.
A 413 and a 429 point to different issues
Anthropic defines 413 request_too_large as exceeding the maximum bytes and points to endpoint limits. Start with the actual route and hosting platform.
429 rate_limit_error is separate: it can mean an organization rate limit, a monthly usage-tier spend cap, or a Claude Code workspace spend limit. It is not a size error.
A 413 below a headline number does not identify the responding component. The reviewed docs do not settle client serialization accounting either, so keep the cause unknown until response origin and route are verified. Preserve any available request ID and sanitized evidence for support instead of retrying blindly.
Collect evidence in a safe diagnostic order
Editorial advice: record the API path, hostname, platform, and endpoint type, then check the matching current limit. The 128 MB announcement is not a substitute. Keep path and hostname separate so a deployment alias does not hide a provider change.
Keep a sanitized status, error, timestamp, request ID or headers if present, SDK version, and known request path. Remove keys, prompts, files, and customer data; mark an unknown response origin as unverified. Avoid saving a full production body when a redacted reproduction is sufficient.
If available, record the client's final serialized body size. Character and token counts are not body bytes. If exact size is unavailable, label the estimate and do not guess an MB conversion or encoding multiplier.
If it appears below cap, inspect client, ingress, and gateway settings and logs one layer at a time, then recheck provider docs. This sequence does not identify the cause by itself. For escalation, share the timestamp, route, cap checked, and redacted error evidence with the relevant platform support team.
Reduce the request before trying again
An identical retry does not shrink the request. Reduce or split content before retrying; unchanged retries are not a size fix.
Remove irrelevant or duplicate images and pages, crop while preserving needed evidence, or split semantically independent work. These suggestions do not guarantee success; preserve required context and data permissions. After splitting, verify that each request still contains the information needed for its own step.
The first-party Files API has a separate 500 MB cap; that does not prove it bypasses Messages limits, and Google-managed Claude does not support Files API. Check feature support; keep request size, context, token quota, and rate or spend limits distinct.
For other authentication, permission, or rate-limit errors, continue with the AI relay troubleshooting guide: https://puppyip.com/en/resources/ai-relay-error-troubleshooting-guide. Identify the response source first. A proxy affects the connection path; it cannot grant account permissions or change platform limits.
Sources
Frequently Asked Questions
Does this also raise file-upload limits in a Claude subscription?
The release note only confirms a request-size change for Messages and Token Counting on the first-party Claude API. It does not establish a change to file-upload rules in Claude subscription products; check the relevant product limits separately.
Does the Files API limit of 500 MB mean PDF page counts are unlimited?
No. That is the request-size cap for a separate endpoint. It does not establish unlimited PDF pages, model processing capacity, or file-format support; check the documentation for the feature you plan to use.