Vals AI MysteryMechanism: leaderboard
MysteryMechanism (Vals AI) — agents rediscover a hidden mathematical relationship by choosing experiments against a noisy oracle under a limited query budget; accuracy over the private task set.
Metric: Accuracy (%). Source: www.vals.ai. Saturation forecast: Around July 2028. 29 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra (Max) | 53.15 | #6 (GPT-6 Astra) |
| 2 | Claude Opus 5.5 (Max) | 49.55 | #3 (Claude Opus 5.5) |
| 3 | Claude Sonnet 5.5 (Max) | 49.1 | #14 (Claude Sonnet 5.5) |
| 4 | Claude Fable 5.1 (Max) | 47.75 | #10 (Claude Fable 5.1) |
| 5 | GPT-6.1 Sol (Max) | 46.4 | #7 (GPT-6.1 Sol) |
| 6 | Gemini 4 Argon (High) | 45.5 | #9 (Gemini 4 Argon) |
| 7 | Claude Opus 5 (Max) | 37.39 | #16 (Claude Opus 5) |
| 8 | Gemini 3.8 Flash (High) | 36.49 | #30 (Gemini 3.8 Flash) |
| 9 | Muse Spark 1.3 (Max) | 36.04 | #31 (Muse Spark 1.3) |
| 10 | Claude Haiku 5.5 (Max) | 35.14 | #43 (Claude Haiku 5.5) |
| 11 | Muse Wonder Full (Max) | 35.14 | #12 (Muse Wonder Full) |
| 12 | GPT-5.6 Sol (Max) | 33.33 | #22 (GPT-5.6 Sol) |
| 13 | Grok 4.6 (High) | 30.63 | #29 (Grok 4.6) |
| 14 | GPT-6 Sol (Max) | 30.18 | #19 (GPT-6 Sol) |
| 15 | Muse Wonder (Max) | 27.48 | #18 (Muse Wonder) |
Interactive version: theaggregate.ai/benchmark?slug=vals-ai-mysterymechanism · How It Works · Data refreshed daily, snapshot 2026-11-05.