Benchmarks · coding
GameDevBench
Uses first-party pass@1 across all 333 game-development tasks. Score: pass@1 across all 333 tasks. Collected by Jake Boggs' benchmark dashboard.
18 models · 20 results · data as of 2026-10-29 · waynechi.com ↗
Leaderboard
Opus 5.5 max
All results
20 rows| # | Model | Effort | GameDevBench |
|---|---|---|---|
| 1 | Opus 5.5 | max | 70.6% |
| 2 | GPT-6.1 Sol | xhigh | 70.0% |
| 3 | GPT-6 Astra | high | 68.8% |
| 4 | Fable 5 | max | 67.3% |
| 5 | GPT-5.6 Sol | xhigh | 63.7% |
| 6 | GPT-5.6 Sol | high | 63.1% |
| 7 | Muse Spark 1.2 | max | 61.0% |
| 8 | GPT-5.6 Sol | medium | 58.6% |
| 9 | Kimi K3 | max | 58.0% |
| 10 | Opus 4.8 | max | 55.9% |
| 11 | GPT-5.5 | xhigh | 54.7% |
| 12 | Gemini 3 Pro | high | 53.8% |
| 13 | GPT-5.4 | xhigh | 52.0% |
| 14 | Gemini 3 Flash | high | 46.8% |
| 15 | GPT-5.4 Mini | xhigh | 43.2% |
| 16 | GLM 5.2 | max | 38.4% |
| 17 | Sonnet 4.5 | high | 34.8% |
| 18 | Kimi K2.5 | max | 20.7% |
| 19 | Haiku 4.5 | xhigh | 18.6% |
| 20 | Qwen 3.5 397B | 5.4% |