FinRiskAtlas - Evidence-Grounded Processing - Information Extraction: leaderboard
Metric: Field-Level Score (%). Source: arxiv.org. 33 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.6 Luna | 85.97 |
| 2 | GPT-5.6 Sol | 85.07 |
| 3 | Seed 2.0 Lite | 84.48 |
| 4 | Kimi K2.5 | 84.28 |
| 5 | GPT-5.4 | 83.98 |
| 6 | GPT-5.4 Mini | 83.78 |
| 7 | GPT-5.6 Terra | 83.48 |
| 8 | Qwen 3.7 Max | 82.79 |
| 9 | GLM-5.2 | 81.69 |
| 10 | Nemotron 3 Ultra | 81.49 |
| 11 | GLM-5 | 81.39 |
| 12 | Qwen 3 235B A22B | 81.25 |
| 13 | DeepSeek V3 (0324) | 79.12 |
| 14 | Qwen 3 14B | 76.82 |
| 15 | Claude Opus 4.7 | 76.62 |
Interactive version: theaggregate.ai/benchmark?slug=finriskatlas-evidence-grounded-processing-information-extraction · How It Works · Data refreshed daily, snapshot 2026-09-19.