- Suite
- V0.4 public
- Models
- 0
- Updated
- Not yet
- Top score
- n/a
Six focused ManimCE tasks
under sandboxed execution
- Source checks
- Render metadata
- Visual sanity checks
- Cost accounting
- Runtime
- Output tokens
Leaderboard
| # | Model | Score | Visual | Cost | Time | Output tokens | Review |
|---|
Overall ManimBench score
Higher is better. Automated checks plus optional visual review.
Run V0.4
V0.4 is file-backed. Generate one ManimCE file per task, place the files in an output folder, then run the launcher.
python -m pip install -e ".[dev,render]"- Write six files under
outputs/<model>/, named with the V0.4 task IDs. python start_benchmark.py- Select the public suite, the model output folder, and the container sandbox for official runs.
Each output file must define one ManimCE MainScene. The default suite expects basic_manim_layout.py, calculus_derivative_graph.py, linear_algebra_transformation.py, geometry_measurement_diagram.py, probability_distribution.py, and advanced_math_explanation.py.