ManiBench: leaderboard
Metric: Executability (%): share of zero-shot generated Manim CE scripts for ManiBench's 12 pilot problems that render without errors or deprecated-API use in a sandbox with a 60-second timeout; temperature 0; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K2.5 | 66.7 | #139 |
| 2 | Claude Sonnet 4 | 66.7 | #194 |
| 3 | Gemini 2.5 Pro | 33.3 | #145 |
| 4 | Qwen 3.5 Plus | 33.3 | #123 |
| 5 | Qwen 3 235B A22B | 25 | #304 |
| 6 | Llama 3.1 8B Instruct | 8.3 | #1018 |
| 7 | Gemma 3 27B | 0 | #596 |
| 8 | Qwen 2.5 Coder 32B Instruct | 0 | #516 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=manibench · How It Works · Data refreshed daily, snapshot 2026-10-11.