FinEval - Multimodal: leaderboard
FinEval's multimodal section: questions over financial charts, candlestick plots, indicators, seals, documents and scenarios, spanning analysis and decision support, risk control and asset allocation. Scored as the README's weighted average accuracy (%) over its tasks; higher is better. Rows the source marks as limited by multi-image input or context window are included as printed.
Metric: Weighted average accuracy (%). Source: github.com. Saturation forecast: Estimated already saturated. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen-VL-Max | 76.3 | #1226 |
| 2 | InternVL3-78B | 72.5 | #733 |
| 3 | InternVL2.5-78B | 71.5 | #1018 |
| 4 | GPT-4o (2024-11-20) | 68.5 | #864 |
| 5 | InternVL3-8B | 65.4 | #1695 |
| 6 | Qwen 2.5 VL 7B | 65.4 | #1796 |
| 7 | Gemini 2.5 Pro (Preview 03-25) | 64.7 | #234 |
| 8 | Claude 3.7 Sonnet (20250219) | 62.9 | #399 |
| 9 | MiniCPM-V-2.6 | 60.1 | #3095 |
| 10 | Llama 3.2 11B Instruct | 50.9 | #5039 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=fineval-multimodal · How It Works · Data refreshed daily, snapshot 2026-11-12.