Skip to main content
BenchLM

Vals MysteryMechanism (MysteryMechanism)

We show this table for reference; we do not rank on it.

Data verified 36 confirmed releases in the last 30 daysFollow model changes

Can agents rediscover sealed mathematical mechanisms through bounded experiments?

MysteryMechanism score on MysteryMechanism — September 27, 2026

We mirror the published mysterymechanism score view for MysteryMechanism. GPT-6 Astra leads the public snapshot at 53.15%, followed by Claude Opus 5.5 (49.55%) and Claude Sonnet 5.5 (49.10%). We do not use these results to rank models overall.

19 modelsAgenticCurrentDisplay onlyUpdated September 27, 2026

MysteryMechanism score table (19 models)

Score
1
GPT-6 AstraOpenAI · Closed
53.15%
2
Claude Opus 5.5Anthropic · Closed
49.55%
3
Claude Sonnet 5.5Anthropic · Closed
49.10%
4
Claude Fable 5.1Anthropic · Closed
47.75%
5
Claude Opus 5Anthropic · Closed
37.39%
6
Gemini 3.8 FlashGoogle · Closed
36.49%
8
GPT-5.6 SolOpenAI · Closed
33.33%
9
Grok 4.6xAI · Closed
30.63%
10
GPT-6 SolOpenAI · Closed
30.18%
11
Grok 4.7xAI · Closed
25.23%
12
Qwen3.8 MaxAlibaba · Open weight
23.87%
13
GLM-5.3Z.AI · Open weight
22.97%
14
MiMo-V2.6-FlashXiaomi · Open weight
21.62%
15
DeepSeek V4.1 FlashDeepSeek · Open weight
21.17%
17
GPT-6 LunaOpenAI · Closed
19.37%
18
MiMo-V2.6-ProXiaomi · Open weight
15.31%
19
GPT-5.6 LunaOpenAI · Closed
14.41%

How MysteryMechanism is shown here

BenchLM mirrors the public Vals AI MysteryMechanism leaderboard captured from https://www.vals.ai/benchmarks/mysterymechanism and updated by Vals on September 27, 2026. The snapshot preserves overall scores, uncertainty, latency, cost-per-test metadata, and task-level scores where Vals publishes them.

MysteryMechanism is display only on BenchLM. Vals proprietary or Vals-hosted aggregate views are useful context, but BenchLM does not use them as weighted ranking inputs or as a replacement for benchmark-native source records.

Snapshot

19 Vals rows1 task viewsprivate datasetTasks: OverallDisplay only

The published MysteryMechanism snapshot places GPT-6 Astra first at 53.15%. The third row is 4.05 points behind. The broader top-10 range is 22.97 points, so the table still separates the published systems.

19 models have been evaluated on MysteryMechanism. The benchmark falls in the Agentic category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. MysteryMechanism is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About MysteryMechanism

Year

2026

Tasks

Sealed mathematical mechanisms probed under an experiment budget

Format

Accuracy score

Difficulty

Agentic scientific discovery

An agent is given a sealed mechanism it can only probe through a bounded budget of experiments, and is scored on how well it recovers the underlying rule. The task set is private and the runner is external, so the board stays display only and outside weighted rankings.

Freshness and provenance

Version

MysteryMechanism 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

Questions

What does MysteryMechanism measure?

Can agents rediscover sealed mathematical mechanisms through bounded experiments?

Which model leads the published MysteryMechanism snapshot?

GPT-6 Astra currently leads the published MysteryMechanism snapshot with 53.15% mysterymechanism score. BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many models are evaluated on MysteryMechanism?

The September 27, 2026 snapshot contains 19 AI models.

Last updated: September 27, 2026 · mirrored from the public benchmark leaderboard

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 2,000+ readers.

One email each week. Unsubscribe anytime.