SDS Field Extraction (Text Pipeline): leaderboard
SDS Field Extraction (2026) scores LLMs on extracting 235 schema fields, including tables and hazard pictograms, from ten industrial chemical Safety Data Sheets against manually validated ground truth. The board is field-level accuracy in percent, averaged over zero-shot, few-shot and chain-of-thought prompts, with each PDF first converted to Markdown; higher is better.
Metric: Mean field-level accuracy (%) of structured extraction from ten industrial Safety Data Sheets (235 schema fields across ten sections, about 50,000 labeled fields) against manually validated ground truth, averaged over zero-shot, few-shot and chain-of-thought prompts, with each PDF first converted to Markdown by PyMuPDF4LLM; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 1.5 Pro | 82 | #1049 |
| 2 | GPT-4o | 80 | #741 |
| 3 | Claude 3.7 Sonnet | 79 | #458 |
| 4 | Llama 3.1 70B Instruct | 69 | #1546 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=sds-field-extraction-text-pipeline · How It Works · Data refreshed daily, snapshot 2026-11-13.