CFMME - Recognition, Detection and Extraction: leaderboard
CFMME (2026) is a Chinese financial multimodal evaluation set for vision-language models, built from financial documents, charts, seals, tables and exam questions. This board is the average in percent of seal recognition, table recognition, formula recognition, seal detection and information extraction scores, tasks on which only non-reasoning models were tested; higher is better.
Metric: Average (%) of seal recognition 1-NED, table recognition TEDS, formula recognition CDM, seal detection mAP and information extraction F1 on CFMME (6,052 Chinese financial multimodal instances over eight financial image types), zero-shot, mean of three runs, the non-reasoning model of each family; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 9 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Qwen 3 VL 235B A22B Instruct | 77.18 | #532 | |
| 2 | GLM-4.5V | Non-reasoning | 71.77 | #754 |
| 3 | Qwen 3 VL 8B Instruct | 71.04 | #1029 | |
| 4 | InternVL3.5-241B-A28B | Non-reasoning | 59.49 | #733 |
| 5 | InternVL3.5-8B | Non-reasoning | 59.03 | #1604 |
| 6 | GPT-4o | 57.89 | #751 | |
| 7 | ERNIE 4.5 VL 28B A3B | 56.61 | #2121 | |
| 8 | Kimi-VL-A3B-Instruct | 50.39 | #2253 | |
| 9 | Llama 3.2 11B Instruct | 27.42 | #5174 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=cfmme-recognition-detection-and-extraction · How It Works · Data refreshed daily, snapshot 2026-11-21.