T2S-Bench - Fault Localization: leaderboard
Metric: Exact match (%) on the 204 Fault Localization questions 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 | 74.51 | #145 |
| 2 | Claude Sonnet 4.5 | 65.2 | #138 |
| 3 | Gemini 2.5 Flash | 64.22 | #237 |
| 4 | GPT-5.2 | 62.75 | #105 |
| 5 | Claude Sonnet 4 (20250514) | 61.27 | #211 |
| 6 | Qwen 3 32B | 53.43 | #424 |
| 7 | GPT-5.1 | 51.47 | #131 |
| 8 | Kimi K2 0905 | 50.98 | #282 |
| 9 | Qwen 3 235B A22B 2507 Instruct | 50 | #291 |
| 10 | Qwen 3 235B A22B 2507 (Thinking) | 50 | #253 (Qwen 3 235B A22B 2507) |
| 11 | Claude Haiku 4.5 (20251001) | 49.51 | #251 |
| 12 | GPT-4.1 Mini | 48.04 | #346 |
| 13 | DeepSeek R1 0528 | 46.57 | #217 |
| 14 | GPT-4o | 44.12 | #333 |
| 15 | DeepSeek V3 (0324) | 43.63 | #332 |
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-fault-localization · How It Works · Data refreshed daily, snapshot 2026-10-11.