Same model, wildly different results: Fable 5.1 shines in Claude Code but stumbles elsewhere
altryne · x · 2026-09-21
altryne compares Fable 5.1 across coding agent harnesses and finds dramatic differences: in Claude Code the model documents requests in Linear, proactively offers to kick off jobs, and opens a PR in 22 seconds — while in other (unnamed) harnesses it behaves like "the slow kid." His takeaway: the harness around a model can matter as much as the model itself. He also laments running out of Fable tokens.
Related event: Hands-On: Fable 5.1 Crushes Astra in Coding Agent Tests(3 posts)→
More from coding & agent
- KDE Drafts AI Policy: Use LLMs, But Don't Tell Anyone — carsonfarmer · 2026-09-21
- 9 practical Jev API recipes: browser agent flies through Google Flights in 7 seconds — alexcovo_eth · 2026-09-21
- GitHub's 17.4k-Star stop-slop Teaches LLMs to Strip AI Writing Tells — tom_doerr · 2026-09-21
- Solo attorney running agent fleet: evidence gates to stop coding-agent regression loops — Specialist_Call_1257 · 2026-09-21
- Code review classification saves 122k tokens, author says even 10% savings is a free win — draginol · 2026-09-21
- Microsoft open-sources IQ Solution Accelerator unifying enterprise data, knowledge and decisions — adnan_hashmi · 2026-09-21