Muse Wonder burns 6.5x GPT-6.1 Sol's tokens on identical benchmark tasks
steipete · x · 2026-10-14
Developer Morgan Linton reports Muse Wonder is extremely token-hungry: his Muse Code weekly limit ran out in under 12 hours at Low Effort. Preliminary VulcanBench Frontier v4 data (19 of 23 tasks) shows:
- Muse Wonder Full averaged 724k effective tokens per task, 3.5x Claude Opus 5.5 (205k) and 6.5x GPT-6.1 Sol (112k), and was the most token-hungry on 18 of 19 tasks
- Mean 18.5 minutes per task vs 4.9 for Sol and 9.0 for Opus
- Functional scores were close (0.933 / 0.966 / 0.934)
Accuracy looks fine, but cost-per-task and speed are currently far worse than rivals, making full benchmarking expensive or months-long.
More from Models
- ChatGPT downweights self-promotional listicles, goes straight to Reddit for reviews — lilyraynyc · 2026-10-14
- First full Gemini 4 Argon hands-on review lands on YouTube: "Gemini is back" — Away_Cold_4943 · 2026-10-14
- Gemini 4 catches most fake API methods in hallucination stress test, one miss remains — haider1 · 2026-10-14
- Bindu Reddy predicts Fable 5.5 and Astra 6.1 launches within two weeks, with price cuts — bindureddy · 2026-10-14
- Pro members fear being left out as Gemini 4 'Argon' rumors swirl — MicahBerkley · 2026-10-14
- Google details safety evaluations and mitigations for Gemini 4 Argon launch — ancadianadragan · 2026-10-14