About FrontierScience
Definition and scoring
- Organisation
- OpenAI
- Category
- Knowledge
- Version
- 2026
- Direction
- higher is better
- Ranking use
- Reference
- Contamination risk
- Unknown
A benchmark for research-level scientific reasoning, designed to separate frontier models on difficult science tasks that mix domain knowledge with deep reasoning. Every genuine source row stays tied to its exact model label, configuration, benchmark version and evaluation system. The summary chart shows one best compatible score per canonical product; Score 2.0 admits only explicitly mapped, frozen protocols and keeps incompatible configurations separate.
Open benchmark source ↗