Frontier Financial Judgement
Finance & Accounting · 2026-07-22
82 realistic cases built with professional equity analysts: given a company primer and news, an agent labels a target article's newness, importance and direction for the stock. All-label accuracy needs all three labels to match the experts.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.5 | 52.4 |
| GPT-5.6 Sol | 51.2 |
| GPT-5.6 Terra | 46.3 |
| GPT-5.6 Luna | 42.7 |
| DeepSeek-V4-Pro | 40.2 |
| GPT-5.4 | 40.2 |
| Opus 4.8 | 39.0 |
| Sonnet 4.6 | 36.6 |
| DeepSeek-V4-Flash | 34.1 |
| MiMo-V2.5 | 32.9 |
| GPT-5.4 Mini | 30.5 |
| Qwen3.6 35B A3B | 22.0 |
| Qwen3.7-Plus (source unspecified) | 20.7 |
| Nemotron 3 Super 120B A12B | 8.5 |