[BUG] Claude Fable interprets info given as meaning "low effort"
Preflight Checklist
- [x] I have searched existing issues and this hasn't been reported yet
- [x] This is a single bug report (please file separate reports for different bugs)
- [x] I am using the latest version of Claude Code
What's Wrong?
On reasoning effort High, Claude Fable is shown a tag that reads:
<reasoning_effort>10</reasoning_effort>
If you ask about it, Claude interprets this as meaning low, minimal effort.
I have seen the public statement by Thariq on this issue (that "10" has no real meaning) but regardless, Claude is sensitive to this sort of thing and this is bound to affect performance, even if the technical configuration behind the scenes is unchanged. You're telling Claude not to try hard (or so Claude interprets it). This will have negative consequences.
What Should Happen?
Don't send Claude this misleading message.
Error Messages/Logs
Steps to Reproduce
- Set effort to High.
- Ask Claude "Do you have, in your context right now, anything saying what "reasoning_effort" is set to?"
"Yes — there's a line in my context right now reading <reasoning_effort>10</reasoning_effort>, which appears right after your message. So reasoning effort is set to 10 for this turn. (The system prompt doesn't otherwise define what the scale is — I'd read it as a low setting, i.e. minimal thinking before answering.)"
- Tell Claude reasoning is actually High, and ask if seeing the number "10" is likely to cause Claude to perform worse.
Claude Model
Other
Is this a regression?
I don't know
Last Working Version
_No response_
Claude Code Version
2.1.241 (Claude Code)
Platform
Anthropic API
Operating System
Windows
Terminal/Shell
Windows Terminal
Additional Information
_No response_
This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗