Grid Game Adaptation (8×8): leaderboard
Long-horizon agent game (Grid Game, Pezeshkpour and Hruschka, Megagon Labs, 2026) on an 8x8 grid: the LLM agent sees a 5x5 window and must gather key fragments and exit through a door before its step budget runs out, while hidden rule tiles, latent cells, observation noise, slipping moves, an energy budget, environment shifts, teleports, spreading hazards and drift in its own action costs are all active. Success rate in percent over 50 random episodes; higher is better.
Metric: Success rate (%; 50 random 8x8 episodes with all perturbations active: collect the key fragments and exit through the door before the step budget runs out). Source: arxiv.org. Saturation forecast: Around February 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash | 42 | #159 |
| 2 | GPT-5.2 | 40 | #172 |
| 3 | Gemini 3 Pro | 28 | #135 |
| 4 | GPT-5 Mini | 22 | #349 |
| 5 | Qwen 3 235B A22B (Non-reasoning) | 0 | #622 (Qwen 3 235B A22B) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=grid-game-adaptation-8-8 · How It Works · Data refreshed daily, snapshot 2026-11-12.