First agent swarm scaling law: Fable 5.2 scales to 0.35 with 100 agents on Lean proofs
infoxiao · x · 2026-11-10
A chart from Anthropic's system card, shared as "the first agent swarm scaling laws," measures scores on formalizing Fermat's Last Theorem-adjacent statements in Lean (60 statements, best within 18 hours, averaged over 3 runs) across agent team sizes.
- Claude Fable 5.2: 0.12 with 1 agent → 0.21 at 10 → 0.24 at 30 → 0.35 at 100, climbing steadily
- Claude Opus 5.5: 0.09 → only 0.21 at 100 agents, markedly weaker scaling
Fable 5.2 isn't just stronger per-agent — its gains from multi-agent scaling are steeper. Caveat: this task uses different theorems than the Lean task in the Opus 5.5 system card, so scores aren't comparable across the two cards.
Related event: 1,000 Fable 5.2 Agents Formalize Fermat's Last Theorem in 30 Hours(20 posts)→
More from Models
- Claude Fable 5.2 scores 64.9% on RoboDojo robot control, dominating rivals — burny_tech · 2026-11-10
- Claude Fable 5.2 video demo wows on pacing, but AI-sounding music persists — jiayuan_jy · 2026-11-10
- Gemini Notebooks keeps pulling context from old conversations, user reports — AnonymousDork929 · 2026-11-10
- User complains model takes four minutes to compare two 16-token completions — cephaloform · 2026-11-10
- DavidAU releases Qwen3.8-40B-Ultimate-Heretic-Uncensored-Turbo-Brilliance-50x-AGI-NEO-GGUF merge — mp3m4k3r · 2026-11-10
- Uncensored Gemma 4 31B 'heretic' build hits Hugging Face, 11.9GB quantized — alexcovo_eth · 2026-11-10