GBENCH One-shot Coding GScore
Games · 2026-04-06
GBENCH One-shot Coding evaluates models by giving them complete descriptions of novel game environments and allowing one code submission with no retries. GScore combines percentile performance across games, successful execution, and a small response-time penalty on a 0-100 scale.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5 | 65.1 |
| GPT-5.6 Sol | 63.7 |
| GPT-5.6 Terra | 61.7 |
| Muse Spark 1.1 | 57.3 |
| GPT-5.5 | 56.2 |
| Opus 4.8 | 54.9 |
| GPT-5.4 | 54.1 |
| Gemini 3.5 Flash | 53.7 |
| Gemini 3.1 Pro Preview | 53.6 |
| Kimi K3 | 51.7 |
| Gemini 3 Pro Preview | 50.8 |
| Qwen3.7-Max | 48.7 |
| Opus 4.6 | 48.6 |
| GPT-5.3-Codex | 48.1 |
| GPT-5.6 Luna | 48.0 |
| Opus 4.7 | 47.8 |
| Sonnet 4.6 | 47.3 |
| GLM-5.2 | 47.2 |
| Grok 4.5 | 47.0 |
| GPT-5.1 | 46.6 |
| Sonnet 5 | 46.1 |
| GPT-5.2 | 45.9 |
| Opus 4.5 | 44.5 |
| MiMo-V2.5-Pro | 43.2 |
| GLM-5.1 | 42.2 |
| Qwen3.6-Max-Preview | 42.1 |
| GPT-5.1-Codex | 41.0 |
| GPT-5.2-Codex | 40.8 |
| Kimi K2.6 | 40.3 |
| GLM-4.7 | 39.6 |
| Kimi K2.7 Code | 39.3 |
| Gemini 2.5 Pro | 39.2 |
| Gemini 3 Flash Preview | 38.8 |
| Gemma 4 26B A4B IT | 37.1 |
| GLM-5 | 37.0 |
| DeepSeek-V4-Pro | 36.4 |
| Grok 4 | 36.2 |
| Grok Build 0.1 | 36.1 |
| Kimi K2.5 | 36.0 |
| MiniMax M3 | 36.0 |