Claude Fable 5.1 Halves Agent Failure Rate in PINNACLE Benchmark

Anthropic's Claude Fable 5.1 halved agent failure rates to 7 per 100 tasks with a 0.7% hallucination rate in the PINNACLE benchmark, completing 3.6x more tasks at 1.8x cost.

2026-09-03 ~ 2026-09-03 · 2 related posts

• PINNACLE: Claude Fable 5.1 halves agent failure rate but costs $2.46 per correct answer — ryanshrout · 2026-09-03
• Fable 5.1 halves agent failure rate to 7% with 0.7% hallucinations, at 1.8x the cost — ryanshrout · 2026-09-03