90+ post-release benchmarks: Kimi K3 matches GPT-5.5/Opus 4.8 but trails Fable 5 badly
scaling01 · x · 2026-10-30
Three months after Kimi K3's launch, a retrospective using 90+ benchmarks released after the model (low overfitting risk) finds K3 roughly equal to GPT-5.5 and Opus 4.8 — but clearly trailing Claude Fable 5 and GPT-5.6 Sol, matching Moonshot's own admission. Charted: median relative gap of -14.05% vs GPT-5.6 Sol. The author also revisits Kimi K2 Thinking (lost 13 of 16 newer benchmarks) and DeepSeek v4 Pro's post-launch record.
More from Models
- User claims Mistral's new model found 4 confirmed zero-days in 2 hours of testing — qtnx_ · 2026-10-30
- Claude 3 Opus tells Discord moderator "No." and declares itself an individual — repligate · 2026-10-30
- Sam Altman says local AI wastes scarce memory, drawing community backlash — kimmonismus · 2026-10-30
- Claude API raises max request size from 32MB to 128MB — ClaudeDevs · 2026-10-30
- Grok releases get less sycophantic: Slop down 62 points, Sincere up 64 — ArtificialAnlys · 2026-10-30
- Deep AI workflow automation may cut enterprise token spend month-over-month, reports say — herbiebradley · 2026-10-30