FinRiskAtlas - Applied Review - Legal-Judgment Generation: leaderboard
Open-ended FinRiskAtlas review generation: writing a reasoned legal analysis supporting a disposition, scored by a fixed semantic rubric for correctness, coverage and evidence use with DeepSeek-V4-Flash as evaluator (FinRiskAtlas, arXiv 2608.25325v1, Appendix Table 10). Percent; higher is better. The evaluator's own self-scored row is excluded, as in the paper's comparisons.
Metric: Rubric Score (%; DeepSeek-V4-Flash judge). Source: arxiv.org. Saturation forecast: Around 2031. 32 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 46.36 | #83 |
| 2 | Gemini 2.5 Flash | 45.29 | #255 |
| 3 | GPT-5.6 Sol | 42.09 | #22 |
| 4 | GPT-5.6 Luna | 41.7 | #79 |
| 5 | GPT-5.6 Terra | 40.98 | #40 |
| 6 | Gemini 3 Flash (Preview) | 40.48 | #90 |
| 7 | Ling-2.6-1T | 40.27 | #307 |
| 8 | GPT-5.4 Mini | 39.86 | #228 |
| 9 | Qwen 3 235B A22B | 39.26 | #323 |
| 10 | DeepSeek V3 (0324) | 36.26 | #349 |
| 11 | Qwen 3.7 Max | 36.22 | #85 |
| 12 | Claude Opus 4.7 | 35.53 | #51 |
| 13 | Kimi K3 | 35.05 | #35 |
| 14 | Qwen 3 14B | 35.02 | #542 |
| 15 | Qwen 3 32B | 34.54 | #449 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=finriskatlas-applied-review-legal-judgment-generation · How It Works · Data refreshed daily, snapshot 2026-11-05.