TruthInsightBench
Explore score history ↗TruthInsightBench evaluates open-ended scientific discovery agents on 40 blind tasks across 10 scientific domains. Each task presents a neutral research objective with frozen data while source conclusions are withheld. A fixed LLM-based judge scores agents' claims along six dimensions with 29 artifact-grounded items, and automated aggregation produces overall scores. Data and scoring code are publicly available.
- Released
- 2026-09-04
- Evaluates
- AI Scientist, General AI, Language & Knowledge
- Openness
- unknown
- Importer
- Claire Radar
- Review status
- ai-reviewed
TruthInsightBench paper and code
No reported scores are on record for this benchmark yet.