Muse Wonder scores 93.4 on Extended NYT Connections benchmark, among priciest tested
zero0_one1 · reddit · 2026-10-13
Lech Mazur's Extended NYT Connections benchmark (940 puzzles) updated: Muse Wonder (high) scored 93.4, ahead of Muse Spark 1.3 (high) at 85.1, and is among the most expensive models tested. The leaderboard's top tier remains GPT-6 Astra xhigh (98.1), Gemini 3.1 Pro Preview (97.7), GPT-6.1 Sol (97.4) and Kimi K3 (97.4); open-weight DeepSeek V4.1 Flash (high) hits 90.0, while Lling 2.6 1T trails at 4.1. The benchmark adds extra trick words and isn't affiliated with NYT.
More from Models
- Claude Opus 5 beats CPAs 100% to 37% on month-end close benchmark, at 2% the cost — McDonaghMatthew · 2026-10-14
- Experiment Wrap: Opus 5.5-Trained 8M Model Generalizes Well but Still Falls Short — Angaisb_ · 2026-10-14
- Gemini 4 Argon nearly doubles AlphaQubit throughput, designs novel enzymes — haider1 · 2026-10-14
- Gemini 4 Argon live on Vercel AI Gateway with 50% off until Dec 1 — steren · 2026-10-14
- Every's vibe check: Gemini 4 Argon puts Google back on the frontier — every · 2026-10-14
- Xbow: AI agents ran a six-day exploit campaign on gpt-5.6-sol, costing ~$40,000 in API fees — rez0__ · 2026-10-14