Our pick for agentic coding changed
- What happened
- We changed our agentic coding pick for the first time since February: Fable 5 replaces Claude Opus 4.8.
- Why it matters
- If you run long multi-step coding tasks, the model that fails less on hour three saves real money and review time.
- What to do
- Move long-horizon agent pipelines to Fable 5; keep Opus 4.8 for review-heavy work.
For the first time since February we are changing our agentic coding recommendation. Fable 5 replaces Claude Opus 4.8 as our pick for long-horizon, multi-step coding tasks.
The deciding data came from our own production pipeline: across two weeks of parallel runs, Fable 5 produced 31% fewer dead-end loops on tasks longer than one hour and recovered from failed test runs without operator intervention noticeably more often.
This is a split verdict, not a dethroning. Opus 4.8 remains our recommendation for review-heavy work — code review, refactoring with strict constraints, and anything where deliberate pacing beats raw persistence.
31% fewer dead-end loops on long-horizon tasks in our pipeline runs. Opus 4.8 stays our pick for review-heavy work.
Affected tools & models
Never need to catch up again
The weekly delta — only verdict changes and act-now items. No digest filler.