GBA Eval Overall
Games · 2026-05-02
Weighted accuracy of an AI-written Game Boy Advance emulator against Mesen2, combining replay/gameplay, audio, and CPU/DMA/memory/timer/IRQ checks.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5 | 79.6 |
| Claude Fable 5 | 74.5 |
| Opus 4.8 | 70.9 |
| Grok 4.5 | 65.4 |
| Sonnet 5 | 65.3 |
| GPT-5.5 | 53.2 |
| GPT-5.6 Sol | 52.6 |
| Sonnet 4.6 | 48.8 |
| Kimi K3 | 48.3 |
| Opus 4.6 | 44.1 |
| Opus 4.7 | 43.8 |
| GPT-5.4 | 31.6 |
| Muse Spark 1.1 | 7.9 |
| Grok Build 0.1 | 2.4 |
| MiniMax M3 | 0.9 |
| Kimi K2.6 | 0.9 |
| Kimi K2.7 Code | 0.9 |
| Gemini 3.1 Pro Preview | 0.8 |
| Qwen3.7-Max | 0.4 |
| GLM-5.2 | 0.0 |
| GLM-5.1 | 0.0 |
| MiniMax M2.7 | 0.0 |