Meta's Muse Wonder nears Claude on benchmarks, but its best versions will be closed
kevinsxu · x · 2026-10-13
kevinsxu reacts to Meta's frontier return: he's been using Muse more and more and is excited for Meta, with benchmark data showing Muse Wonder Full (max) matching or beating Claude and GPT models on several evals (DeepSWE 79.0, SWE-Marathon 57.5, ProgramBench 59.0).
The concern is open source: reports say Muse Wonder's best versions will be 'closed', sold directly to businesses, pitting Meta against Anthropic and OpenAI for corporate AI contracts. He warns this is bad news for US open-source AI — NVIDIA plus the labs it will sooner or later acquire may be the only game left. Meta shares rose after the Muse launch.
More from Companies & People
- Puentes flies 15 top Latin American engineers to SF to compete with the best — yenkel · 2026-10-13
- Google DeepMind Robotics is hiring a pretraining & data quality research scientist in London — bousmalis · 2026-10-13
- Zuck's watermelon meme is his answer to calls to slow down frontier AI — beffjezos · 2026-10-13
- PyTorch to host "From Academia to Adoption" workshop on Oct 19 in San Jose — PyTorch · 2026-10-13
- Ethan Mollick's new book teased ahead of James Cham fireside chat — DynamicWebPaige · 2026-10-13
- a16z resurfaces Altman's 2015 take: chronically negative hires are a startup's biggest hidden mistake — a16z · 2026-10-13