Mollick: Muse Wonder impresses, but Meta's thinking model trails the Big Three
emollick · x · 2026-10-13
Ethan Mollick shared early impressions of Meta's new models: Muse Wonder looks good in limited testing, while Muse Spark Thinking is "fine but really doesn't match the current Big Three," with odd language and tone and loose facts. More striking is his drowned neogothic city twigl shader test comparison since April, showing Meta's remarkably fast progress in a short time.
More from Models
- Muse agent completes three.js boat task, then leaves Matthew Berman a baffling message — MatthewBerman · 2026-10-13
- Muse senior engineer benchmark results pulled over flawed run, retest underway — danshipper · 2026-10-13
- "Zero chance" a Meta model competes with Google, OpenAI or Anthropic right now — iruletheworldmo · 2026-10-13
- Dev turns EmbeddingGemma 2 into an on-device STT model, transcribing 1 hour of audio in 13 seconds — solyarisoftware · 2026-10-13
- Alexandr Wang hails Theo's model rankings: Opus 5.5 and Muse Wonder alone in S tier — alexandr_wang · 2026-10-13
- Reddit user rails against Gemini's decline: wrong bulb model, 13 failed script fixes — hotboxuzi · 2026-10-13