FinBen - FinNum: leaderboard
MultiFinBen v3 Table 3 standardized score on Greek FinNum (grfinnum), the Greek-language financial numeral-understanding task. Source: the official Open-FinLLM static Space.
Metric: Normalized Score. Source: huggingface.co. Saturation forecast: Estimated already saturated. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 4 Scout Instruct | 49.12 | #539 |
| 2 | Llama 3.1 70B Instruct | 46.34 | #561 |
| 3 | Qwen 2.5 32B Instruct | 36.77 | #498 |
| 4 | O3 Mini | 20.98 | #272 |
| 5 | GPT-4o | 9.18 | #343 |
| 6 | DeepSeek V3 | 7.43 | #321 |
| 7 | Qwen2.5-Omni-7B | 0.4 | #624 |
| 8 | Gemma 3 27B (IT) | 0 | #523 |
| 9 | Gemma 3 4B (IT) | 0 | #992 |
| 10 | Qwen2-Audio-7B-Instruct | 0 | #880 |
| 11 | DeepSeek-R1-Distill-Qwen-32B-Japanese | 0 | #638 |
| 12 | SALMONN-7B | 0 | #1293 |
| 13 | SALMONN-13B | 0 | #1260 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=finben-finnum · How It Works · Data refreshed daily, snapshot 2026-10-20.