AstaBench - Literature Understanding: leaderboard
Agents find papers (PaperFindingBench, LitQA2-FullText-Search), answer long-form questions with citations (ScholarQA-CS2), build literature-review tables (ArxivDIGESTables-Clean) and answer full-text multiple-choice questions (LitQA2-FullText); the score is the weighted mean of the five, the two LitQA2 tasks at half weight. Unlike LitQA2 alone it mixes retrieval, synthesis and table building in one average.
Metric: Weighted mean of 5 literature tasks, the two LitQA2 tasks at half weight (0-1). Source: allenai-asta-bench-leaderboard.hf.space. Saturation forecast: Around May 2027. 23 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 (xHigh) | 0.61 | #79 (Claude Opus 4.7) |
| 2 | GPT-5.4 (2026-03-05) (xHigh) | 0.58 | #87 (GPT-5.4 (2026-03-05)) |
| 3 | GPT-5 (Medium) | 0.55 | #167 (GPT-5) |
| 4 | O3 (2025-04-16) (Medium) | 0.48 | #216 (O3 (2025-04-16)) |
| 5 | GPT-4.1 | 0.46 | #495 |
| 6 | Claude Sonnet 4 (20250514) (Thinking) | 0.46 | #387 (Claude Sonnet 4 (20250514)) |
| 7 | GPT-5 Mini (2025-08-07) (Medium) | 0.37 | #343 (GPT-5 Mini (2025-08-07)) |
| 8 | Claude 3.5 Haiku (20241022) | 0.36 | #1578 |
| 9 | Gemini 2.5 Flash (Preview 05-20) | 0.33 | #426 |
| 10 | GPT-4o (2024-08-06) | 0.32 | #740 |
| 11 | Llama 4 Scout Instruct | 0.2 | #1543 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=astabench-literature-understanding · How It Works · Data refreshed daily, snapshot 2026-11-13.