[Bug] Rate limit error despite low API usage when using Opus 4.6 (1m)

Status Fixed / completed
Reported on v2.1.49
Maintainer reply ✓ Yes — ashwin-ant
Activity 9 comments · opened Feb 20, 2026 · closed Apr 19, 2026
💡 Likely answer: A maintainer (ashwin-ant, collaborator) responded on this thread — see the highlighted reply below.

Bug Description

"Rate limit reached"; nowhere near usage limit however. Sessions are not particularly long-running. This has impacted all open sessions, not just any single one. I was using Opus 4.6 (1m Context). Seems to have been resolved by switching to Opus 4.6 (Default).

Environment Info

  • Platform: darwin
  • Terminal: ghostty
  • Version: 2.1.49
  • Feedback ID: 4e319a3c-d4e6-4dbb-856b-dc513b4f27b2

Errors

[{"error":"Error: NON-FATAL: Lock acquisition failed for /Users/accipio/.local/share/claude/versions/2.1.49 (expected in multi-process scenarios)\n    at PFR (/$bunfs/root/claude:2599:2098)\n    at iyA (/$bunfs/root/claude:2599:1209)\n    at processTicksAndRejections (native:7:39)","timestamp":"2026-02-20T16:22:42.723Z"},{"error":"Error: 429 {\"type\":\"error\",\"error\":{\"type\":\"rate_limit_error\",\"message\":\"Extra usage is required for long context requests.\"},\"request_id\":\"req_011CYKdSgBj9rFLwf9ck8Hph\"}\n    at generate (/$bunfs/root/claude:262:28456)\n    at makeRequest (/$bunfs/root/claude:301:5435)\n    at processTicksAndRejections (native:7:39)","timestamp":"2026-02-20T16:22:48.419Z"}]

View original on GitHub ↗

9 Comments

github-actions[bot] · 6 months ago

Found 3 possible duplicate issues:

  1. https://github.com/anthropics/claude-code/issues/26918
  2. https://github.com/anthropics/claude-code/issues/26496
  3. https://github.com/anthropics/claude-code/issues/26575

This issue will be automatically closed as a duplicate in 3 days.

  • If your issue is a duplicate, please close it and 👍 the existing issue instead
  • To prevent auto-closure, add a comment or 👎 this comment

🤖 Generated with Claude Code

puchkovvolodymyrgm · 6 months ago

Same, in 10 mins of slow coding, reach 90% of 5 hour limit

somanyusers · 5 months ago

here too. Issue appears specific to bogus Opus rate limit enforcement in Claude Code, not actual usage overage. Claude Code v2.1.76, macOS, Claude Pro (OAuth).

Claude Code consistently returns "API Error: Rate limit reached" for all Opus 4.6 requests, despite actual usage well below limits (usage-dashboard-confirmed). Web ui Opus 4.6 works fine on same account. Fresh auth (logout/login) doesn't resolve it, switching to Sonnet 4.6 via /model sonnet works around the bug.

shehwar-ahmad · 5 months ago

Same issue.

korn007 · 5 months ago

Same issues what happen

chok · 5 months ago

In my case, it was a too long session. I removed it and it works again

AnnNulee · 5 months ago

Same here. Max 5x, Opus 4.6.
Earlier today, I was implementing a single feature — hit 58% in just 40 minutes. I assumed I'd been inefficient with file operations, but it kept climbing and maxed out at 100% shortly after. Waited the full 3-hour cooldown.
Started a new session just now — fixed a single merge conflict in a JSON file (read, edit conflict markers, validate). Maybe 7-8 turns, minimal tool usage. Already at 5%. Same CLAUDE.md, same project, same model. The drain rate is completely inconsistent with actual workload. The lack of per-turn token visibility makes it impossible to even diagnose whether this is a metering bug or a silent cost increase. Exposing input/output token counts per API call — even just in a local log — would go a long way.

ashwin-ant collaborator · 4 months ago

This was fixed in v2.1.88 — Fixed misleading "Rate limit reached" message when the API returned an entitlement error (e.g. 1M context without Extra Usage) — now shows the actual error with actionable hints (/extra-usage or /model to switch). If you're still seeing this in the latest version, please comment with your version and repro and we'll reopen.

github-actions[bot] · 4 months ago

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.