Benchmarks
347 sources · 1,258 models × 3,038 benchmarks · 158,912 scoresAi2 scientific-agent suite covering literature, code execution, data analysis, and end-to-end discovery. Container for test-set components under the April 2026 scoring update; the overlapping suite and category composites are not imported. ScholarQA-CS2 and end-to-end discovery were rescored with replacement judges. Coverage includes thirteen identified ReAct baselines, with their source-linked harness versions preserved.
No signal or difficulty estimate. Held out of the fit by design.
This is a benchmark-family page. Scores are recorded on the component benchmarks below rather than directly on the parent, because the components use different score scales and directions.
| Metric | Models | Leader | Direction | Date |
|---|---|---|---|---|
| ArxivDIGESTables-Clean (AstaBench 2026) | 13 | Opus 4.70.4 | higher is better | — |
| CORE-Bench-Hard (AstaBench 2026) | 13 | GPT-5.40.9 | higher is better | — |
| DiscoveryBench (AstaBench 2026) | 13 | GPT-5.50.4 | higher is better | — |
| DS-1000 (AstaBench 2026) | 13 | Gemini 3.1 Pro Preview0.8 | higher is better | — |
| E2E-Bench (AstaBench 2026) | 13 | Opus 4.70.7 | higher is better | — |
| E2E-Bench-Hard (AstaBench 2026) | 13 | Opus 4.70.7 | higher is better | — |
| LitQA2-FullText (AstaBench 2026) | 13 | Opus 4.60.9 | higher is better | — |
| LitQA2-FullText-Search (AstaBench 2026) | 13 | Opus 4.60.9 | higher is better | — |
| PaperFindingBench (AstaBench 2026) | 13 | Opus 4.70.4 | higher is better | — |
| ScholarQA-CS2 (AstaBench 2026) | 13 | GPT-5.40.9 | higher is better | — |
| SUPER-Expert (AstaBench 2026) | 13 | Opus 4.70.4 | higher is better | — |