Paperbench Expanded
Category: AI for ResearchPaper reproduction beyond machine learning: 93 high-difficulty tasks across 12 research areas and six scientific domains, run in the field's own software and graded by clean replay. The strongest agent fully reproduces 13 of them.
Pricing
Free
Free (Open Source / Free Tier)
Prices change often. Confirm current plans on the official website before you buy.
Listed features
- What we kept from PaperBench, and what we expanded
- What a task looks like
- How a paper becomes a task
- Difficulty by research area
- What separates partial progress from a reproduction
- Data release