SDABench: leaderboard
SDABench (Southeast University, ECNU and HKUST, 2026) tests scientific data analysis on held-out synthetic datasets from biology, chemistry, environment, geography and physics, organized by six capability types with programmatically derived answers. The score is open-ended answer accuracy in percent, averaged over the six types; higher is better.
Metric: Accuracy (%) of open-ended answers on the held-out SDA-Synth test split (2,400 instances, 400 per task type across five scientific domains, macro-averaged over the six task types; zero-shot, JSON reasoning and answer fields, the answer scored by relative-error numeric, exact or structured matching with a GPT-4o fallback judge); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 58.33 | #123 |
| 2 | Gemini 3.1 Pro (Preview) | 56.88 | #94 |
| 3 | Claude Sonnet 4.6 | 55.79 | #161 |
| 4 | GLM-5 | 51.67 | #260 |
| 5 | Qwen 3.5 397B A17B | 51.58 | #246 |
| 6 | DeepSeek V3.2 | 51.17 | #406 |
| 7 | Kimi K2.5 | 50.29 | #256 |
| 8 | DeepSeek R1 | 42.13 | #473 |
| 9 | Qwen 3 235B A22B 2507 Instruct | 40.46 | #619 |
| 10 | GPT-5.4 Mini | 37.67 | #389 |
| 11 | Llama 3.3 70B Instruct | 6.96 | #1357 |
| 12 | Llama 3.1 8B Instruct | 5.04 | #4246 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=sdabench · How It Works · Data refreshed daily, snapshot 2026-11-13.