LuminaBench - GameDev: leaderboard
The rating of one playable kart-racing game built by the model in its coding agent, from a rubric gameplay assessment or the site owner's relative placement, none blinded; a build with no working game takes the floor rating of 800. One build per configuration.
Metric: Owner rating of one mario-kart-v1 build. Source: luminabench.com. Saturation forecast: Around January 2028. 10 models tracked. Updated 16 Nov 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Claude Fable 5.2 | High | 1118 | #1 |
| 2 | Claude Opus 5.5 | Max | 1100 | #5 |
| 3 | Claude Sonnet 5.5 | Max | 1097 | #19 |
| 4 | GPT-6 Astra | Max | 1069 | #6 |
| 5 | GPT-6.1 Sol | Max | 1058 | #10 |
| 6 | Gemini 4 Argon | High | 1056 | #12 |
| 7 | GPT-6.1 Terra | Max | 1052 | #21 |
| 8 | Grok 4.7 | xHigh | 1015 | #47 |
| 9 | Claude Haiku 5.5 | High | 997 | #70 |
| 10 | Gemini 3.8 Flash | Low | 800 | #36 |
No result here: #2 Gemini 4 Carbon.
Interactive version: theaggregate.ai/benchmark?slug=luminabench-gamedev · How It Works · Data refreshed daily, snapshot 2026-11-16.