FinRiskAtlas - Applied Review - Decision-View Generation: leaderboard
Open-ended FinRiskAtlas review generation: writing a recommended review opinion for the question presented to the reviewer, 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 February 2027. 32 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V3 (0324) | 71.31 | #349 |
| 2 | Gemini 2.5 Flash | 65.27 | #255 |
| 3 | Gemini 3 Flash (Preview) | 64.61 | #90 |
| 4 | Kimi K3 | 64.13 | #35 |
| 5 | GPT-5.4 | 62.91 | #83 |
| 6 | Kimi K2.5 | 62.74 | #152 |
| 7 | GPT-5.6 Sol | 62.25 | #22 |
| 8 | Seed 2.0 Lite | 58.73 | #161 |
| 9 | GLM-5.2 | 58.59 | #80 |
| 10 | Qwen 3.7 Max | 57.84 | #85 |
| 11 | GLM-5 | 57.16 | #149 |
| 12 | Ling-2.6-1T | 55.69 | #307 |
| 13 | GPT-5.6 Terra | 54.72 | #40 |
| 14 | DeepSeek V4 Pro | 54.48 | #112 |
| 15 | Qwen3-Next-80B | 53.1 | #321 |
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-decision-view-generation · How It Works · Data refreshed daily, snapshot 2026-11-05.