视觉 / 多模态大模型排行榜:图片理解、OCR、图表问答能力评分,附 API 价格与上下文。 · 数据更新于 2026-11-17
选视觉模型时,除了分数,还要看图片分辨率限制、单次支持几张图、中文 OCR 效果和输出价格。表格里的价格是文本 token 的最低输入价,图片 token 的计费方式请以模型页为准。
| 排名 | 模型 | 厂商 | 榜单分 | 最低输入价 | 上下文 | 渠道 |
|---|---|---|---|---|---|---|
| 1 | Claude Fable 5 | Anthropic | 1309±7 | ¥66.79 | 1.00M | 7 |
| 2 | Claude Opus 4.7 | Anthropic | 1300±7 | ¥33.4 | 1.00M | 8 |
| 3 | Qwen 3.8 (Max 0902) | Alibaba | 1299±7 | ¥13.36 | 1.00M | 2 |
| 4 | Claude Opus 4.6 | Anthropic | 1299±6 | ¥33.4 | 1.00M | 6 |
| 5 | Claude Opus 4.7 | Anthropic | 1298±7 | ¥33.4 | 1.00M | 8 |
| 6 | Claude Opus 5.5 | Anthropic | 1295±10 | ¥26.72 | 1.00M | 11 |
| 7 | Muse Spark 1.3 | Meta | 1294±9 | ¥8.35 | 1.05M | 2 |
| 8 | Gemini 3.7 Flash | 1294±9 Preliminary | ¥2.5 | 1.05M | 7 | |
| 9 | Muse Spark 1.2 | Meta | 1294±14 | ¥8.35 | 1.05M | 1 |
| 10 | Claude Opus 4.6 | Anthropic | 1293±6 | ¥33.4 | 1.00M | 6 |
| 11 | Gemini 3.8 Flash | 1293±8 Preliminary | ¥2.5 | 1.05M | 7 | |
| 12 | GPT 6.1 Sol | OpenAI | 1293±11 | ¥6.68 | 1.05M | 9 |
| 13 | Muse Spark 1.3 | Meta | 1291±8 | ¥8.35 | 1.05M | 2 |
| 14 | Gemini 3 Pro Image 2k (nano-banana-pro) | 1289±7 | ¥13.36 | 131K | 2 | |
| 15 | Claude Fable 5.1 | Anthropic | 1289±9 | ¥66.79 | 1.00M | 5 |
| 16 | GPT 5.6 Sol | OpenAI | 1288±7 | ¥13.36 | 1.05M | 8 |
| 17 | Muse Spark 1.3 | Meta | 1288±8 | ¥8.35 | 1.05M | 2 |
| 18 | Claude Opus 4.8 | Anthropic | 1286±6 | ¥33.4 | 1.00M | 11 |
| 19 | Claude Opus 5 | Anthropic | 1285±7 | ¥33.4 | 1.00M | 11 |
| 20 | GPT 5.5 | OpenAI | 1285±6 | ¥16.7 | 1.05M | 7 |
| 21 | GPT 6 Astra | OpenAI | 1283±9 | ¥33.4 | 1.05M | 8 |
| 22 | GPT 5.5 | OpenAI | 1283±6 | ¥16.7 | 1.05M | 7 |
| 23 | GPT 5.4 | OpenAI | 1282±6 | ¥8.35 | 1.05M | 7 |
| 24 | Gemini 3.5 Flash | 1281±7 | ¥5.01 | 1.05M | 7 | |
| 25 | Gemini 3.5 Flash | 1281±7 | ¥5.01 | 1.05M | 7 | |
| 26 | Grok 4.5 | xAI | 1280±7 | ¥13.36 | 500K | 4 |
| 27 | Gemini 3.1 Pro Preview | 1280±5 | ¥6.68 | 1.05M | 6 | |
| 28 | Gemini 3.6 Flash | 1280±8 | ¥2.5 | 1.05M | 7 | |
| 29 | Muse Spark 1.1 | Meta | 1279±7 | ¥8.35 | 1.05M | 1 |
| 30 | Claude Opus 4.8 | Anthropic | 1279±6 | ¥33.4 | 1.00M | 11 |
先用能力分筛出前 20 名,再按场景过滤:发票/合同偏 OCR,图表问答偏推理,截图 Agent 偏工具调用。中文 OCR 建议优先看国产多模态模型。