Meta launches Muse Wonder frontier agent models, tops SWE-Marathon at 57.5%
acbuller · x · 2026-10-13
Meta has unveiled Muse Wonder and Muse Wonder Full, a new family of frontier models targeting personal and professional agentic tasks.
- Muse Wonder Full jumped to 1st place on SWE-Marathon with 57.5%, per evals supported by Abundant Labs on long-horizon coding.
- The published chart shows Muse Wonder Full (xhigh) at 57.5% and Muse Wonder (max) at 53.1%, ahead of Claude Opus 5.5 (max) at 46.3%, Claude Fable 5.1 (max) at 45.6%, and GPT-6 Astra (max) at 42.5%.
- The launch was announced personally by Meta leadership as state of the art for agentic work.
Related event: Meta Unveils Muse Wonder Frontier Model for Personal Agents(81 posts)→
More from coding & agent
- NVIDIA open-sources SMLIB, its NURBS-based CAD kernel, after 20-year partnership — Promptmethus · 2026-10-13
- Readme team launches Restless: self-improving APIs with live logs and agent recovery — testingcatalog · 2026-10-13
- LLMs are System 2: Austin LangChain's framing for pairing fast judgment with slow reasoning — colinmcnamara · 2026-10-13
- A feed of next actions is the only sane way to manage massively parallel agents — _AustinCalvert_ · 2026-10-13
- Grok agent's human-like 'ack' behavior impresses in hands-on test, still trails Claude Code for coding — doooyle · 2026-10-13
- chessmod lets you play rated chess bots inside Claude Code while it works — bushibuilds · 2026-10-13