Sonnet 4.5·4 1M Beta Ends
Key point
Anthropic has ended the 1M context beta for Sonnet 4·4.5 and now blocks requests exceeding 200K.
Details
For Sonnet 4.5 and Sonnet 4, the context-1m-2025-08-07 beta header ended at UTC midnight, and requests exceeding 200K tokens now return a 400 error.
The fix is simple.
- Change the model string to
claude-sonnet-4-6. On this model, 1M context is GA, so the beta header is unnecessary. - Remove the beta header from requests.
- The long-context surcharge is also gone. The 2x premium was already abolished in March.
The likely failure points are clear.
- Cases where
if context > 200K-style branching no longer routes to the 1M path - Long-running sessions where accumulated conversation history exceeds 200K tokens
- Agentic workflows where tool-call and reflection logs pile up quickly
If you're running a gateway, you need to re-check the context limits per model, and setting a hard cap in proxies like Bifrost and LiteLLM can block Anthropic's 400 errors upstream.
The key point is that when production traffic breaks, the priority is to change the model string.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.