Step Game (Lechmazur): leaderboard
Tests spatial reasoning - models must track relative positions of characters over 10 steps (e.g., 'A is to the left of B'). TrueSkill ranking from multi-game tournaments across 75+ models.
Metric: TrueSkill μ. Source: github.com. Saturation forecast: Around December 2026. 75 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 (Medium) | 5.49 | #98 (GPT-5) |
| 2 | O3 (Medium) | 5.32 | #124 (O3) |
| 3 | GPT-5.1 (Medium) | 5.29 | #135 (GPT-5.1) |
| 4 | Gemini 3 Pro (Preview) | 5.03 | #66 |
| 5 | O1 | 4.14 | #183 |
| 6 | Gemini 2.5 Flash | 3.99 | #245 |
| 7 | Gemini 2.5 Pro | 3.82 | #152 |
| 8 | Grok 4.1 Fast (Reasoning) | 3.78 | #217 (Grok 4.1 Fast) |
| 9 | Gemini 2.5 Pro (Preview 03-25) | 3.72 | #123 |
| 10 | DeepSeek V3.2 Exp | 3.67 | #234 |
| 11 | O3 Mini (High) | 3.64 | #272 (O3 Mini) |
| 12 | Grok 4 | 3.47 | #181 |
| 13 | O3 Mini (Medium) | 3.38 | #272 (O3 Mini) |
| 14 | Grok 3 Mini (High) | 3.24 | #281 (Grok 3 Mini) |
| 15 | DeepSeek R1 0528 | 3.21 | #222 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=step-game-lechmazur · How It Works · Data refreshed daily, snapshot 2026-10-20.