xAI Grok 4.6 Model Card - 3DCodeBench: leaderboard
3DCodeBench (Gao et al., 2026) evaluates agentic procedural 3D modeling via code: agents author engine-ready 3D assets through software APIs and are graded for executability and shape fidelity. xAI Grok 4.6 model card (revised 2026-08-17), section 4.2 chart; reward in percent, higher is better; Grok runs in the Grok Build harness.
Metric: Reward (%). Source: media.x.ai. Saturation forecast: Around 2029. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Grok 4.6 (High) | 54 | #29 (Grok 4.6) |
| 2 | Claude Opus 5 (Max) | 49.9 | #16 (Claude Opus 5) |
| 3 | Grok 4.5 (High) | 49.8 | #54 (Grok 4.5) |
| 4 | Claude Opus 4.8 (Max) | 47 | #39 (Claude Opus 4.8) |
| 5 | Claude Fable 5 (Max) | 43.7 | #20 (Claude Fable 5) |
| 6 | Claude Sonnet 5 (Max) | 39.2 | #65 (Claude Sonnet 5) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=xai-grok-4-6-model-card-3dcodebench · How It Works · Data refreshed daily, snapshot 2026-11-05.