Mythos 5.2 tops BenchBench-Cloning at 35.1%; Fable 5.2 refused every task
nlarusstone · x · 2026-11-11
The BenchBench-Cloning leaderboard shows Claude Mythos 5.2 (LSVP) on top at 35.1% pass rate—a big jump from Mythos 5.1's 16.0% though only marginal over GPT-6 Astra's 30.6%. GPT-6.1 Sol scored 22.2%, Claude Opus 5.5 20.5%, Gemini 4 Argon 18.8%. Biology cloning is progressing slowly. Notably, Fable 5.2 didn't make the chart because it refused all tasks; Grok 4.7 (10.9%) and Kimi K3 (8.8%) trail.
More from Models
- Gemini Pro user's screenshot shows a locked '4 Carbon' model tier in the UI — SR_RSMITH · 2026-11-11
- xAI dumps JAX, rewrites full RL/inference stack in C/C++ as Grok 4.8 enters RL — ivan_bezdomny · 2026-11-11
- Goodfire's Curt Tigges: individual LLM neurons are far more interpretable than thought — The Cognitive Revolution · 2026-11-11
- Power User Burned 20-40B Tokens: Calls Gemini 4 Carbon His Favorite Knowledge-Work Model — rseroter · 2026-11-11
- Gemini 4 Carbon Developer Guide: When to Pick Carbon, Thinking Levels & Verify Loops — rseroter · 2026-11-11
- Gemini 4 Carbon spotted on GCI chart scoring ~134, above Gemini 4 Argon — scaling01 · 2026-11-11