GameCraft-Bench - Content Depth: leaderboard
A rubric category of GameCraft-Bench (2026), where coding agents build complete Godot games from natural-language specifications: content depth (enough levels, entities and progression), scored by a multimodal judge from replayed gameplay, in percent. Games that do not launch or submit no playable demonstration score zero; higher is better.
Metric: Content Depth score (%): rubric score for content depth (enough levels, entities and progression), on the 140 GameCraft-Bench tasks (15 game families; complete Godot projects built from natural-language game specifications): a project that fails to launch or submits no parsable demonstration trace scores zero; otherwise the verifier replays the agent's mouse and keyboard traces into gameplay video and frames and a GPT-5.5 multimodal judge scores them against hidden rubric items; higher is better. Source: arxiv.org. Saturation forecast: Around December 2027. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 (High) | 39.48 | #79 (Claude Opus 4.7) |
| 2 | GPT-5.5 (High) | 38.61 | #49 (GPT-5.5) |
| 3 | Kimi K2.6 | 28.07 | #178 |
| 4 | MiMo-V2.5-Pro | 22.59 | #279 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=gamecraft-bench-content-depth · How It Works · Data refreshed daily, snapshot 2026-11-13.