Benchmark

多模态模型能力榜

vision multimodal 多语言

多模态模型能力榜:多维度评测综合成绩。评测分数为事实数据,整理自公开评测体系,仅供学习研究参考。

Top 10

榜单概览

1Qwen3.7 Plus73.02Gemini 3.1 Pro Preview72.63Qwen3.5 397B-A17B72.34GPT-5.571.15Kimi K2.670.66Doubao-Seed-2.0-Pro (hi…69.47SenseNova-6-7-Flash-Alp…69.08GLM-5V-Turbo62.89Step 3.7 Flash61.510Gemma 4 31B IT56.3
数据说明

分数为 0-100 百分制综合成绩(Elo 榜为相对评分,仅供同榜内比较)。评测分数属事实数据,整理自公开评测体系与厂商公开资料,部分分数为厂商自报口径,不同榜单的评测方法与数据集可能存在差异,请结合多个榜单交叉参考。

Ranking

模型排名

名次 模型 机构 主分 综合(含 Agent) 感知 空间 信息图 推理 创作 智能体
1 Qwen3.7 Plus Alibaba Cloud / Qwen Team 73.0 80.5 71.0 52.0 82.0 84.4 64.0 45.3
2 Gemini 3.1 Pro Preview Google 72.6 77.1 71.0 52.5 78.0 83.8 66.5 27.3
3 Qwen3.5 397B-A17B Alibaba Cloud / Qwen Team 72.3 77.9 73.0 52.5 80.0 82.5 63.1 34.0
4 GPT-5.5 OpenAI 71.1 79.6 53.0 54.0 74.0 86.3 73.4 50.7
5 Kimi K2.6 Moonshot AI 70.6 75.6 60.0 52.0 78.0 81.9 69.8 30.0
6 Doubao-Seed-2.0-Pro (high) ByteDance 69.4 75.9 71.0 50.5 75.0 78.1 63.4 39.3
7 SenseNova-6-7-Flash-Alpha SenseTime 69.0 — 60.0 53.5 67.0 86.9 59.7 —
8 GLM-5V-Turbo Zhipu AI 62.8 67.7 47.0 49.5 68.0 76.9 58.4 29.3
9 Step 3.7 Flash StepFun 61.5 66.2 52.0 43.0 62.0 74.4 63.4 28.0
10 Gemma 4 31B IT Google 56.3 59.2 48.0 43.0 55.0 72.5 46.6 17.3
11 Claude Opus 4.8 Anthropic 54.1 61.8 44.0 40.5 60.0 57.5 65.3 46.0
12 Ovis2.6-80B-A3B Alibaba Cloud / Qwen Team 53.7 — 42.0 37.5 53.0 67.5 54.5 —
13 Mistral-Medium-3.5-128B Mistral AI 53.6 56.0 38.0 39.5 56.0 63.8 60.8 14.0
14 MiniMax-M3 MiniMax 51.6 58.3 31.0 39.0 57.0 59.4 63.8 40.0
15 Valley3-32B-Think ByteDance 40.2 — 45.0 36.5 45.0 35.6 43.4 —
16 Llama-4-Scout-17B-16E-Instruct Meta 36.1 — 32.0 31.0 36.0 36.3 45.3 —
17 SAIL-VL2-8B ByteDance 34.7 — 33.0 35.5 37.0 31.3 40.1 —
18 MiniCPM-V-4.6-Thinking OpenBMB 33.6 — 31.0 39.5 36.0 31.3 32.3 —
19 NVIDIA-Nemotron-Nano-12B-v2-VL-BF16 NVIDIA 29.8 — 29.0 30.5 34.0 23.1 38.9 —
20 LLaVA-OneVision-2-8B-Instruct LMMS-Lab 28.8 — 34.0 32.5 28.0 22.5 33.4 —
21 LFM2.5-VL-1.6B Liquid AI 23.3 — 24.0 30.5 22.0 18.8 26.0 —
22 Flash-VL-2B-Dynamic-ISS Meituan 22.2 — 16.0 33.5 16.0 18.8 30.4 —