T2S-Bench: leaderboard
Metric: Exact match (%) over all four question categories on T2S-Bench-MR (500 expert-checked multiple-choice questions over text-structure pairs from scientific papers in six domains; single- and multiple-answer items, only the text and the question given), deterministic decoding at temperature 0; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 45 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 81.4 | #145 |
| 2 | Claude Sonnet 4.5 | 76.8 | #138 |
| 3 | Claude Sonnet 4 (20250514) | 75.6 | #211 |
| 4 | Gemini 2.5 Flash | 72.2 | #237 |
| 5 | GPT-5.2 | 71.8 | #105 |
| 6 | Qwen 3 32B | 69.4 | #424 |
| 7 | GPT-5.1 | 67.8 | #131 |
| 8 | Claude Haiku 4.5 (20251001) | 67.4 | #251 |
| 9 | Kimi K2 0905 | 67 | #282 |
| 10 | Qwen 3 235B A22B 2507 Instruct | 62.8 | #291 |
| 11 | GPT-4o | 61.8 | #333 |
| 12 | Gemini 2.0 Flash | 61.2 | #331 |
| 13 | Qwen 3 235B A22B 2507 (Thinking) | 60.8 | #253 (Qwen 3 235B A22B 2507) |
| 14 | DeepSeek V3 (0324) | 60.6 | #332 |
| 15 | DeepSeek V3.1 | 60.4 | #260 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=t2s-bench · How It Works · Data refreshed daily, snapshot 2026-10-11.