Leaked Fable 5.2 eval table: tops 9 of 13 benchmarks, RLI hits 41.3%
FinanceYF5 · x · 2026-11-10
Part 3 of the same unverified leak, with a full Claude Fable 5.2 eval table:
- Across 13 benchmarks vs Opus 5.5, GPT-6 Astra, Gemini 4 Carbon, Fable 5.2 ranks first on roughly 9
- Highlights: Agentic coding 49.1%, Agentic SWE v2 69.9%, RLI 41.3%, Humanity's Last Exam 74.2%, Robotics 64.9%
- Safety evals: CyberSecureBench 76.3%, CVE Rediscovery 89.6%, with Opus 5.5 baselines marked "without safeguards"
- Claimed a tier above Opus 5.5 on nearly all capabilities, up to 30% faster than 5.1
Related event: Claude Fable 5.2 Benchmarks Leak, Sparking Hoax Debate(18 posts)→
More from Models
- ARLoop: fully-open 1.2B looped transformer beats Huginn-3.5B and Qwen3.5-2B on all 8 benchmarks — hhsun1 · 2026-11-10
- Fable 5.2 first hands-on: model thinks for 2+ minutes without emitting a single token — op7418 · 2026-11-10
- Fable 5.2 excels at chip design and biology but Anthropic nerfed RSI tasks — omarsar0 · 2026-11-10
- AI beats Getting Over It in superhuman fashion, with route evolution logs — flowersslop · 2026-11-10
- Fable 5.2 hits 64.9% on RoboDojo, beating purpose-built robot AI systems — CyberRobooo · 2026-11-10
- Editors have ~1 year before AI writing models truly get taste — and everyone hears the same Claude sound — ChrisGPT · 2026-11-10