Benchmark RadarAI BENCHMARK PROFILE
ResearchClawBench
ResearchClawBench evaluates autonomous scientific research agents across 40 tasks from 10 domains, each grounded in a published paper with provided literature and raw data. It uses expert-curated multimodal rubrics to score target-paper-level re-discovery.
- Released
- 2026-05-28
- Readiness
- Runnable
- Primary field
- General AI
Why it matters
Autonomous research agents claim to accelerate science, but their end-to-end capability is unverified. ResearchClawBench provides a standardized evaluation frontier to measure progress toward reliable research re-discovery.
Motivation
AI coding agents are increasingly used for scientific work, but their end-to-end autonomous research capability remains difficult to verify.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.