Game2Real-Bench - Toy Game - Constraint Reasoning: leaderboard
Game2Real-Bench - Toy Game - Constraint Reasoning. Main paper zero-shot protocol; separate toy/real populations; real-world F1 where applicable. Reported metric: Accuracy (%). Primary published table, with source configuration boundaries preserved.
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Around January 2027. 25 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol | 81.3 | #18 |
| 2 | Claude Opus 4.8 (Thinking) | 62 | #38 (Claude Opus 4.8) |
| 3 | Gemini 3.1 Pro (Preview) | 56.7 | #56 |
| 4 | Gemma 4 31B (IT) | 42.3 | #172 |
| 5 | Qwen 3.8 27B | 39.3 | #96 |
| 6 | Qwen 3.5 27B | 34.1 | #183 |
| 7 | Qwen 3.5 35B A3B | 33.4 | #250 |
| 8 | Gemma 4 12B (IT) | 25.6 | #356 |
| 9 | Qwen 3.5 9B | 24.3 | #367 |
| 10 | Qwen 3.5 4B | 23.6 | #463 |
| 11 | Step3 VL 10B | 21.3 | #477 |
| 12 | Qwen 3.5 2B | 18.7 | #1109 |
| 13 | Qwen 3 VL 30B A3B Instruct | 16.4 | #375 |
| 14 | Qwen 3 VL 8B Instruct | 12.8 | #414 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #7 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=game2real-bench-toy-game-constraint-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-16.