MysteryMechanism (Vals AI) leaderboard
- GPT-6 Astra leads
- 53.1%
- models measured
- 25
- lead over #2
- +3.6 pts
- median
- 27.5%
Leader on MysteryMechanism (Vals AI) · read October 22
On October 22, 2026, 25 of our catalog's models had a MysteryMechanism (Vals AI) result, and OpenAI GPT-6 Astra led with 53.1%, ahead of Anthropic Claude Opus 5.5 at 49.5%.
Agents rediscover a hidden mathematical mechanism by experiment. The score as Vals AI publishes it.
Role in the index
MysteryMechanism (Vals AI) is shown as published and does not feed any of our indexes. Read the methodology
| # | Model | Result | Cost per test | Duration, as printed | Evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|---|---|
| 1 | GPT-6 AstraOpenAI | 53.1% | $1.56 | 8m 09s | 85.1 | $50.00 | |
| 2 | Claude Opus 5.5Anthropic | 49.5% | $4.62 | 19m 38s | 88.8 | $20.00 | |
| 3 | Claude Sonnet 5.5Anthropic | 49.1% | $3.45 | 23m 30s | 87.3 | $10.00 | |
| 4 | Claude Fable 5.1Anthropic | 47.8% | $5.63 | 17m 41s | 84.2 | $50.00 | |
| 5 | GPT-6.1 SolOpenAI | 46.4% | $0.28 | 7m 17s | 82.4 | $10.00 | |
| 6 | Gemini 4 ArgonGoogle | 45.5% | $4.59 | 24m 55s | 81.0 | $10.00 | |
| 7 | Claude Opus 5Anthropic | 37.4% | $3.28 | 16m 53s | 83.0 | $25.00 | |
| 8 | Gemini 3.8 FlashGoogle | 36.5% | $1.73 | 5m 22s | 61.6 | $3.75 | |
| 9 | Claude Haiku 5.5Anthropic | 35.1% | $0.54 | 13m 25s | 70.7 | $0.50 | |
| 10 | GPT-5.6 SolOpenAI | 33.3% | $0.96 | 8m 43s | 73.5 | $20.00 | |
| 11 | Grok 4.6xAI | 30.6% | $1.31 | 27m 55s | 65.7 | $6.00 | |
| 12 | GPT-6 SolOpenAI | 30.2% | $0.46 | 6m 13s | 76.0 | $10.00 | |
| 13 | Muse WonderMeta | 27.5% | $3.39 | 18m 17s | 71.9 | $20.00 | |
| 14 | Grok 4.7xAI | 25.2% | $1.84 | 17m 33s | 66.7 | $6.00 | |
| 15 | Qwen3.8 Max (0902)Qwen | 23.9% | $1.31 | 37m 49s | 62.4 | $6.00 | |
| 16 | GLM 5.3Z.ai | 23.0% | $1.53 | 30m 42s | 70.6 | $4.40 | |
| 17 | Ember-1Fireworks | 22.5% | $1.29 | 20m 15s | 52.6 | $15.00 | |
| 18 | MiMo-V2.6-FlashXiaomi | 21.6% | $0.06 | 35m 08s | 54.7 | $0.28 | |
| 19 | DeepSeek V4.1 FlashDeepSeek | 21.2% | $0.16 | 8m 04s | 62.5 | $0.84 | |
| 20 | Gemini 3.1 Pro PreviewGoogle | 20.7% | $1.07 | 6m 00s | 32.3 | $12.00 | |
| 21 | GPT-6 LunaOpenAI | 19.4% | $0.03 | 6m 01s | 57.3 | $0.50 | |
| 22 | Step 5 PreviewStepFun | 18.5% | $0.60 | 29m 20s | 61.1 | $2.70 | |
| 23 | MiMo-V2.6-ProXiaomi | 15.3% | $0.15 | 44m 00s | 65.4 | $0.87 | |
| 24 | GPT-5.6 LunaOpenAI | 14.4% | $0.11 | 8m 13s | 58.0 | $1.20 | |
| 25 | Mistral Large 4Mistral AI | 12.2% | $2.10 | 35m 37s | 56.4 | $4.18 |
Only models in our catalog are listed, and each appears once with its best published run. 137 rows were read from the source on October 22, 2026, and rows that match no catalog model are left out. Cost per test is what Vals AI reports one test cost on the run behind the result, in US dollars. Every other column is a figure the board publishes for the same run.