Opus 4.6 (1M context) produces garbled terminal output and usage policy API error

Resolved 💬 3 comments Opened Apr 9, 2026 by daniloapr Closed May 24, 2026

Description

When using Claude Code with the Opus 4.6 (1M context) model, asking a question about a recently implemented feature caused the terminal to output garbled/corrupted characters instead of a normal response. Shortly after, the model returned a usage policy API error. Switching to Sonnet 4.6 resolved the issue.

Steps to Reproduce

  1. Set model to Opus 4.6 (1M context) via /model
  2. Ask a question about a feature you just implemented
  3. Observe garbled output in terminal

Observed Behavior

Terminal printed corrupted output instead of a response:

^ n =...".( n"="_~gntrm n...."]-.....)n

  prefr]3~,n...."] .. . .[....r-,..]_ sqmmuler ceker    [ \+]_  1 v    1
  v.. .]11 v.. .]11
   9
   A y t . . . n s t  r c t r e s
  f o r  (2).    . . 2 , n . . .
  Acknowledgement-strip.
  (vaarrrrt

Followed by:

API Error: Claude Code is unable to respond to this request, which appears to violate our Usage Policy.

Expected Behavior

A normal text response to the question.

Environment

  • OS: macOS Darwin 25.3.0
  • Terminal: iTerm2 and native Mac Terminal.
  • Model: Opus 4.6 (1M context)
  • Network: Switched from airplane WiFi to hotel WiFi during session (possible contributing factor)

Workaround

Switching to Sonnet 4.6 (/model claude-sonnet-4-20250514) resolves the issue.

View original on GitHub ↗

This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗