# ExplorationBench AlienLogic Best@3 (ExplorationBench AlienLogic)

> Tests whether a system can discover altered logic rules and use them to prove or correctly decline unseen theorems.

Canonical page: https://benchlm.ai/benchmarks/explorationbench-alienlogic

- Category: [Agentic](/agentic)
- Last updated: October 15, 2026 snapshot

## About ExplorationBench AlienLogic

- Year: 2026
- Tasks: 70 held-out theorems; 24 altered rules
- Format: Held-out accuracy (Best@3)
- Difficulty: Learning and applying unfamiliar inference rules
- Paper: [ExplorationBench: Measuring AI Systems' Exploration in Verifiable Alien Worlds](https://arxiv.org/abs/2609.30199v1)

A proof checker grades 70 held-out goals after four exploration rounds; 25 goals are unprovable and require refusal. Best@3 selects one complete run, with mean and worst results retained separately. This synthetic logic world is distinct from AlienCode. Private code and tasks keep the table outside overall and category rankings.

ExplorationBench AlienLogic is currently displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.

## Leaderboard (10 systems)

| Rank | System | Reasoning | Before exploration | Best@3 | Mean@3 | Worst |
|---|---|---|---|---|---|---|
| 1 | [Grok 4.6](/models/grok-4-6) | xhigh | 42.9% | 81.0% | 72.1% | 65.7% |
| 2 | [Claude Opus 5](/models/claude-opus-5) | max | 43.3% | 79.5% | 75.9% | 69.5% |
| 3 | [Qwen3.8-Max](/models/qwen3-8-max) | max | 45.2% | 77.6% | 74.3% | 71.9% |
| 4 | [GPT-5.6 Sol](/models/gpt-5-6-sol) | max | 51.0% | 76.7% | 75.1% | 72.4% |
| 5 | [Hy4 preview](/models/hy4-preview) | high | 35.2% | 76.2% | 69.2% | 61.4% |
| 6 | [DeepSeek-V4-Pro](https://explorationbench.com/) | max | 33.8% | 75.2% | 66.3% | 61.4% |
| 7 | [Kimi K3](https://explorationbench.com/) | max | 45.2% | 74.3% | 69.4% | 61.4% |
| 8 | [DeepSeek-V4.1-Flash](/models/deepseek-v4-1-flash) | max | 51.4% | 73.8% | 60.6% | 52.9% |
| 9 | [Seed2.1 Pro](/models/seed-2-1-pro) | high | 37.6% | 69.0% | 59.8% | 53.3% |
| 10 | [Gemini 3.8 Flash](/models/gemini-3-8-flash) | high | 45.7% | 59.5% | 59.0% | 58.6% |

Before exploration and Best@3 come from the same selected run. Mean@3 and Worst cover all three runs; each checkpoint averages three answers per task.

Scores follow the current web table. The September 24 paper describes the protocol but reports an earlier AlienCode maximum of 87.6% versus 89.0% on the website. Code and held-out tasks remain private.

Sources: [Official leaderboard](https://explorationbench.com/), [published result data](https://explorationbench.com/assets/js/results.js), and [paper and protocol](https://arxiv.org/abs/2609.30199v1).

## FAQ

### What does ExplorationBench AlienLogic measure?

Tests whether a system can discover altered logic rules and use them to prove or correctly decline unseen theorems.

### Which model leads the published ExplorationBench AlienLogic snapshot?

Grok 4.6 currently leads the published ExplorationBench AlienLogic snapshot with a score of 81.0%.

### How many models are evaluated on ExplorationBench AlienLogic?

The October 15, 2026 snapshot contains 10 systems at their published reasoning settings.

### Does ExplorationBench AlienLogic affect BenchLM's overall score?

Not directly. ExplorationBench AlienLogic is still displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.