Opus/Sonnet 1M context variants broken on Max 20x — misleading 'Rate limit reached' error
Status Closed — not planned
Maintainer reply None cached
Workaround ✓ Mentioned in thread ↓
Activity 11 comments · opened Mar 2, 2026 · closed May 23, 2026
Bug Description
The 1M context model variants (opus[1m] / claude-opus-4-6[1m]) selected via /model fail with "API Error: Rate limit reached" despite ample quota remaining. This has been broken for over a week. The error is misleading — it's not a rate limit issue.
The non-1M variants (Default, explicit claude-opus-4-6) work fine with the same account and quota.
Reproduction
Tested systematically by editing settings.json and launching fresh sessions:
| "model" value in settings.json | Context Window | Result |
|---|---|---|
| (key absent — Default) | 200k | Works |
| "claude-opus-4-6" | 200k | Works |
| "opus[1m]" (via /model menu option 2) | 1M | Rate limit error |
| "claude-opus-4-6[1m]" (manual or rewritten by Claude Code) | 1M | Rate limit error |
Steps
- Run Claude Code v2.1.63
/model→ select "Opus (1M context)" (option 2)- Settings written as
"model": "opus[1m]" - Send any message →
API Error: Rate limit reached /model→ select "Default" → works immediately
Additional observations
- The
/modelmenu itself hints at the issue: selecting the staleclaude-opus-4-6[1m]shows: "Newer version available · select Opus for Opus 4.6" — but selecting "Opus (1M context)" from the menu gives the same broken ID - Session JSONL confirms: failed requests show
"model":"<synthetic>"with"error":"rate_limit"(never reaches a real model), while Default requests show"model":"claude-opus-4-6"with"service_tier":"standard"
Usage at Time of Error
Current session: 5% used
Current week: 19% used (all models)
Sonnet only: 4% usedNowhere near any limit.
Impact
- Users who previously selected a 1M context model are permanently broken with no obvious fix
- The "Rate limit reached" error is misleading — users wait for a reset that never helps
- The only workaround is using "Default" (200k context) or manually setting
"model": "claude-opus-4-6"and not using/model
Environment
- Claude Code: v2.1.63
- Plan: Claude Max 20x
- Platform: Windows 11 Pro (win32)
- Terminal: Windows Terminal with PowerShell 7
- Shell (Claude Code): bash
- Update channel: latest
11 Comments
Confirmed on macOS (arm64)
Same behavior reproduced on macOS:
| Detail | Value |
|---|---|
| Claude Code | v2.1.63 |
| Platform | macOS Tahoe 26.3 (Darwin 25.3.0, arm64) |
| Machine | MacBook Pro M2 |
| Terminal | macOS Terminal / zsh |
| Plan | Claude Pro 20x (same account as original report) |
Reproduction
/model→ select Opus (1M context) → settings written asopus[1m]API Error: Rate limit reached/model→ select Default → works immediatelyIdentical to the Windows behavior. This is not platform-specific — it's a server-side or client model-routing issue affecting the
[1m]context variants across all platforms.@cadfan Did you manage to find other workarounds? Default doesn't work for me either
@yulonglin The most likely cause is a stale
[1m]model value persisted in your settings — this overrides/modelpicker selections, which could explain why Default doesn't work for you either.Check your
settings.json(run/configor look in~/.claude/settings.json). If there's a"model"key set to anything with[1m]in it, remove that line entirely and restart Claude Code.If there's no
"model"key in your settings and Default still fails, that's likely a different issue. In that case, worth checking:/status)claude --version)Mine is probably a different issue. I was using a 5x Max account, and resolved it by logging out and logging back in again. Restarting sessions didn't help. I never activated or saw in my settings 1m models, just "opus" or nothing at all (when I'm using the default one)
Confirmed on Linux (WSL2) via VSCode extension
| Detail | Value |
|---|---|
| Claude Code | VSCode extension (Opus 4.6 1M) |
| Platform | Linux 6.6.87.2-microsoft-standard-WSL2 |
| Terminal | VSCode integrated terminal / bash |
| Plan | Claude Max |
Same behavior — switched to Opus 4.6 (1M context) via
/model, immediately gotAPI Error: Rate limit reachedon the next message. Switching back to Default works fine. Usage was nowhere near any limit.The error message is misleading — makes you think you've hit a quota when it's actually a model routing issue with the
[1m]variants.I am on linux, Claude 2.1.68 and 20x Max plan. The 1M models don't even show as an option.
Max user here:
why can't I choose the 200k??? and why is 1m broken all of the sudden???????!!!! HOW TO FIX?!
Still hitting this as of April 2026. Running on
claude-opus-4-6[1m], any skill/subagent that specifiesmodel: sonnetin its frontmatter immediately fails with "API Error: Rate limit reached". This has nothing to do with actual rate limits — it's been broken for over a month now since the 1M context feature shipped. The 1M model variants seem to inherit into skill/subagent model overrides, making the entire skill/plugin ecosystem unusable when on a 1M context session. This is a significant regression that's been open for far too long.AS OF TODAY THIS IS RESOLVED 💯
I'm not even sure neither Claude or Anthropic Engineers are reading those issues where we shout in the void.
But i'm not getting those error in 2.1.97
Closing for now — inactive for too long. Please open a new issue if this is still relevant.
This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.