Benchmarks · reasoning
MysteryMechanism
Can agents rediscover sealed mathematical mechanisms through bounded experiments?
21 models · 21 results · updated 2026-10-29 · vals.ai ↗
Leaderboard
GPT-6 Astra max
Heavy line: best score at each price. Thin lines join one model's effort levels. Cost uses each model's median subscription plan, or API price with no plan data. 10 more models under All models in settings.
More from this benchmark
1 published measuresTime per task29 results · lower is better
10 more models under All models in settings.
All results
21 rows| # | Model | Effort | MysteryMechanism | Subscription $ / task | API $ / task | Time |
|---|---|---|---|---|---|---|
| 1 | GPT-6 Astra | max | 53.2% | $0.067 | $1.56 | 8.1 min |
| 2 | Opus 5.5 | max | 49.5% | $0.070 | $4.62 | 20 min |
| 3 | Sonnet 5.5 | max | 49.1% | $0.052 | $3.45 | 24 min |
| 4 | Fable 5.1 | max | 47.7% | $0.88 | $5.63 | 18 min |
| 5 | GPT-6.1 Sol | max | 46.4% | $0.017 | $0.28 | 7.3 min |
| 6 | Gemini 4 Argon | high | 45.5% | $3.27 | $4.59 | 25 min |
| 7 | Opus 5 | max | 37.4% | $0.057 | $3.28 | 17 min |
| 8 | Gemini 3.8 Flash | high | 36.5% | $0.070 | $1.73 | 5.4 min |
| 9 | Muse Spark 1.3 | max | 36.0% | $0.43 | $0.85 | 7.5 min |
| 10 | Haiku 5.5 | max | 35.1% | $0.011 | $0.54 | 13 min |
| 11 | GPT-5.6 Sol | max | 33.3% | $0.027 | $0.96 | 8.7 min |
| 12 | Grok 4.6 | high | 30.6% | $0.13 | $1.31 | 28 min |
| 13 | GPT-6 Sol | max | 30.2% | $0.014 | $0.46 | 6.2 min |
| 14 | Grok 4.7 | xhigh | 25.2% | $0.17 | $1.84 | 18 min |
| 15 | Qwen 3.8 Max | max | 23.9% | $0.15 | $1.31 | 38 min |
| 16 | GLM 5.3 | max | 23.0% | $0.16 | $1.53 | 31 min |
| 17 | MiMo V2.6 Flash | high | 21.6% | free | $0.057 | 35 min |
| 18 | DeepSeek V4.1 Flash | high | 21.2% | $0.012 | $0.16 | 8.1 min |
| 19 | GPT-6 Luna | max | 19.4% | $0.002 | $0.026 | 6.0 min |
| 20 | MiMo V2.6 Pro | max | 15.3% | $0.13 | $0.15 | 44 min |
| 21 | GPT-5.6 Luna | max | 14.4% | $0.005 | $0.11 | 8.2 min |