Muse Wonder Full 1.0 is hard to benchmark: $50/mo plan hits weekly limits after a few runs
PawelHuryn · x · 2026-10-14
Pawel Huryn's update on benchmarking Muse Wonder Full 1.0: on a $50/mo subscription he hit weekly limits after several runs; the "contributor" toggle doesn't seem to slow allowance burn, there's no API or known pricing. Earlier: Sonnet 5.5 (max) makes 7x more turns than Astra (verified multiple times) but is the slowest and priciest in his benchmark. More scores coming.
More from Models
- Snap launches as first fully sub-quadratic LLM in production with constant compute per token — ryanallred · 2026-10-14
- xAI says legacy SuperGrok Heavy plans can't be restored after billing merge mistake — RexDouglass · 2026-10-14
- HyperThink uses a hypernetwork to replace long chain-of-thought with one parameter update — Donggyun Kim · 2026-10-14
- Uberti: background voice cancellation is underparameterized for noisy audio — juberti · 2026-10-14
- Gemini 4 Argon leaps in 3D reconstruction: 81.2 on BuildingBench at ~1/5 cost — Lianhuiq · 2026-10-14
- WebRTC author Justin Uberti: diarization is the wrong path, train LLMs on noise — juberti · 2026-10-14