FinRiskAtlas - FinRisk-Ask - Conditional Request Alignment: leaderboard
Conditional request alignment (CRA) in FinRisk-Ask: mean three-level alignment of evidence requests with expert-verified needs, computed only over recorded Ask states where the model chose to ask, judged by DeepSeek-V4-Flash (FinRiskAtlas, arXiv 2608.25325v1, Appendix Table 11). Percent; higher is better; the evaluator's self-scored row is excluded.
Metric: CRA (%; DeepSeek-V4-Flash request-alignment judge). Source: arxiv.org. Saturation forecast: Around February 2027. 32 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.7 Max | 87.67 | #85 |
| 2 | Kimi K3 | 86.61 | #35 |
| 3 | Claude Opus 4.7 | 85.85 | #51 |
| 4 | Gemini 3 Flash (Preview) | 83.55 | #90 |
| 5 | Gemini 2.5 Flash | 83.51 | #255 |
| 6 | GPT-5.6 Terra | 82.58 | #40 |
| 7 | Ling-2.6-1T | 82.58 | #307 |
| 8 | GPT-5.4 | 82.13 | #83 |
| 9 | Nemotron 3 Ultra | 82.12 | #245 |
| 10 | GPT-5.6 Luna | 81.95 | #79 |
| 11 | GLM-5.2 | 81.58 | #80 |
| 12 | GPT-5.6 Sol | 81.27 | #22 |
| 13 | Kimi K2.5 | 79.97 | #152 |
| 14 | DeepSeek V4 Pro | 79.7 | #112 |
| 15 | DeepSeek V3 (0324) | 79.22 | #349 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=finriskatlas-finrisk-ask-conditional-request-alignment · How It Works · Data refreshed daily, snapshot 2026-11-05.