GameCraft-Bench: leaderboard
GameCraft-Bench: agents build complete Godot games across 140 tasks; overall rubric score in percent. Each named agent, model and explicit reasoning-effort configuration is preserved; no best-effort pooling.
Metric: Overall rubric score (%). Source: tongxuluo.github.io. Saturation forecast: Around June 2027. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5 (xHigh) | 68.44 | #12 (Claude Opus 5) |
| 2 | Claude Fable 5 (High) | 65.72 | #16 (Claude Fable 5) |
| 3 | GPT-5.6 Sol (High) | 60.5 | #19 (GPT-5.6 Sol) |
| 4 | Kimi K3 | 56.96 | #30 |
| 5 | Claude Opus 4.7 (High) | 41.46 | #49 (Claude Opus 4.7) |
| 6 | DeepSeek V4 Flash (0731) | 40.61 | #78 |
| 7 | GPT-5.5 (High) | 39.49 | #31 (GPT-5.5) |
| 8 | GLM-5.2 | 39.12 | #73 |
| 9 | Kimi K2.6 | 30.65 | #104 |
| 10 | MiMo-V2.5-Pro | 24.1 | #158 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=gamecraft-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.