MysteryMechanism

Rediscovering hidden mathematical mechanisms through a limited number of experiments. Run by Vals AI.

CategoryReasoning
MetricAccuracy (%)
SourceVals AI (evaluator)
LicenseNone stated
Benchmark pagewww.vals.ai
StatusScored
Weight0.9 within Reasoning
FitDiscrimination 1.25, noise 0.32

Results

#ModelProviderResultDate
1 GPT-6 Astra Max OpenAI 53.2% 2026-11-09
2 Claude Fable 5.2 High Anthropic 51.8% 2026-11-09
3 Claude Opus 5.5 Max Anthropic 49.6% 2026-11-09
4 Claude Sonnet 5.5 Max Anthropic 49.1% 2026-11-09
5 Claude Fable 5.1 Max Anthropic 47.7% 2026-11-09
6 GPT-6.1 Sol Max OpenAI 46.4% 2026-11-09
7 Gemini 4 Argon High Google 45.5% 2026-11-09
8 Claude Opus 5 Max Anthropic 37.4% 2026-11-09
9 Gemini 3.8 Flash High Google 36.5% 2026-11-09
10 Muse Spark 1.3 Max Meta 36.0% 2026-11-09
11 Claude Haiku 5.5 Max Anthropic 35.1% 2026-11-09
12 Muse Wonder Full Max Meta 35.1% 2026-11-09
13 GPT-5.6 Sol Max OpenAI 33.3% 2026-11-09
14 Grok 4.6 High xAI 30.6% 2026-11-09
15 GPT-6 Sol Max OpenAI 30.2% 2026-11-09
16 Muse Wonder Max Meta 27.5% 2026-11-09
17 Kimi K3.1 Pro Max Moonshot AI 26.1% 2026-11-09
18 Grok 4.7 Extra High xAI 25.2% 2026-11-09
19 Qwen3.8 Max Alibaba 23.9% 2026-11-09
20 GLM-5.3 Max Zhipu AI 23.0% 2026-11-09
21 Ember-1 Fireworks 22.5% 2026-11-09
22 Beam 1.0 Reflection 22.1% 2026-11-09
23 MiMo-V2.6-Flash Xiaomi 21.6% 2026-11-09
24 DeepSeek V4.1 Flash High DeepSeek 21.2% 2026-11-09
25 Gemini 3.1 Pro Preview High Google 20.7% 2026-11-09
26 GPT-6 Luna Max OpenAI 19.4% 2026-11-09
27 Step 5 Preview StepFun 18.5% 2026-11-09
28 MiMo-V2.6-Pro Xiaomi 15.3% 2026-11-09
29 GPT-5.6 Luna Max OpenAI 14.4% 2026-11-09
30 Mistral Large 4 High Mistral 12.2% 2026-11-09