MWS Vision Bench: leaderboard
MTS AI's MWS Vision Bench evaluates Russian-language document understanding across full-page OCR, structured parsing, key-information extraction, grounding, and reasoning VQA on 1,302 business-document samples; scored by overall average.
Metric: Overall Score (%). Source: huggingface.co. Saturation forecast: Around March 2027. 46 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Fable 5 | 79.9 | #16 |
| 2 | GPT-5.6 Sol | 79 | #19 |
| 3 | GPT-5.5 | 77.8 | #31 |
| 4 | Kimi K3 | 77.7 | #30 |
| 5 | GPT-5.6 Terra | 74.9 | #41 |
| 6 | Kimi K2.6 | 74.3 | #104 |
| 7 | Claude Sonnet 5 | 73.1 | #57 |
| 8 | GPT-5.6 Luna | 72.7 | #68 |
| 9 | Gemini 3.6 Flash | 72.3 | #51 |
| 10 | Qwen 3.7 Plus | 70.8 | #92 |
| 11 | Claude Opus 4.6 | 70.4 | #63 |
| 12 | Qwen 3.6 35B A3B | 69.4 | #226 |
| 13 | Qwen 3.6 27B | 69.1 | #170 |
| 14 | Gemini 2.5 Pro | 69 | #152 |
| 15 | Kimi K2 | 68.6 | #243 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mws-vision-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.