Kimi K3 and Fable 5 show nearly identical failure patterns on a software benchmark
FinanceYF5 · x · 2026-07-21
A comparison chart shows Kimi K3 and Fable 5 failing in very similar ways on a software engineering benchmark. About 65% of failures are near-misses for both models, and both largely preserve existing baselines, with regression rates of 11% for Kimi and 10% for Fable.
The accompanying analysis says the two models have a per-task correlation of 0.72, the highest cross-vendor similarity on record. It also notes that no task showed a full 4/4 success vs 0/4 failure split in either direction, which may indicate the benchmark is nearing saturation.
Related event: Kimi K3 Matches Fable 5 in SWE Benchmarks at a Third of the Cost(7 posts)→
More from Models
- GPT-6.1 Terra arrives, trained for fast and efficient computer use — mark_k · 2026-10-27
- Nat Lambert: narrow RL training on internal infra is a poor fit for open-weight models — mervenoyann · 2026-10-27
- Leaked Terminal-Bench 4.0 chart shows GPT-6.1 Terra leading at lowest cost, unverified — Angaisb_ · 2026-10-27
- GPT-6.1 Terra hits Astra-level agent scores at 3x speed and one-tenth the cost — OpenAIDevs · 2026-10-27
- OpenAI brings GPT-6.1 Terra to the API and Codex for fast, cheap computer use — OpenAIDevs · 2026-10-27
- GPT-6.1 Terra rolls out to all Plus, Pro, Enterprise and Edu users — OpenAI · 2026-10-27