Curvebench Env
Frontier
CurveBench: Vision-language model evaluation for hierarchical tree structure extraction from images
- Domain
- rl-env
- License
- mit
- Published
- Feb 2026
Cite
Notes
Only stored in your browser.
Top score 73.1% by Gemini 3.1 Pro Preview - 15 models reporting (5 frontier)
Score history
10Top models
15
Related tools
1Implementations, trainers, datasets and scaffolds linked to this eval.
FAQ
- What is Curvebench Env?
- CurveBench: Vision-language model evaluation for hierarchical tree structure extraction from images
- What is the current top score on Curvebench Env?
- The top reported score is 73.1% by Gemini 3.1 Pro Preview, across 15 models reporting (5 from frontier labs).
- How can a model improve its Curvebench Env score?
- Tools linked to Curvebench Env on Sophon include Curvebench ENV RL Env (Community) - RL environments, datasets, and scaffolds that target this eval.
- What license is Curvebench Env under?
- Curvebench Env is available under mit.