MMBench
Indexed results from MMBench. Each ranking retains its own metric and evaluation conditions.
nature_relation
mmBench · MMBench · nature_relation · percent · Higher is better
- OpenGVLab/InternVL2-40Bopengvlab93.5%
- Qwen: Qwen2.5 VL 72B Instructqwen93.5%
- NCSOFT/VARCO-VISION-2.0-14Bncsoft91.3%
- YannQi/R-4Byannqi91.3%
- Google: Gemma 3 27Bgoogle90.2%
- Qwen/Qwen2.5-VL-7B-Instructqwen89.1%
- OpenGVLab/InternVL2_5-8Bopengvlab88%
- OpenGVLab/InternVL3-8Bopengvlab88%
- Qwen/Qwen2-VL-72B-Instructqwen88%
- DeepSeek: DeepSeek-VL2deepseek85.9%
- OpenGVLab/InternVL3-2Bopengvlab85.9%
- Qwen/Qwen2.5-VL-32B-Instructqwen85.9%
image_emotion
mmBench · MMBench · image_emotion · percent · Higher is better
- OpenGVLab/InternVL2-40Bopengvlab84.4%
- OpenGVLab/InternVL2_5-4Bopengvlab83.3%
- Qwen/Qwen2-VL-72B-Instructqwen83.3%
- Qwen/Qwen2-VL-7B-Instructqwen83.3%
- OpenGVLab/InternVL2_5-8Bopengvlab82.2%
- OpenGVLab/InternVL3-8Bopengvlab82.2%
- Qwen: Qwen2.5 VL 72B Instructqwen82.2%
- ibm-granite/granite-vision-3.2-2bibm-granite81.1%
- Google: Gemma 3 27Bgoogle80%
- HuggingFaceM4/Idefics3-8B-Llama3huggingfacem480%
- ibm-granite/granite-vision-3.1-2b-previewibm-granite80%
- NCSOFT/VARCO-VISION-2.0-14Bncsoft80%
physical_property_reasoning
mmBench · MMBench · physical_property_reasoning · percent · Higher is better
- OpenGVLab/InternVL2-40Bopengvlab79.7%
- Qwen/Qwen2-VL-72B-Instructqwen77.2%
- NCSOFT/VARCO-VISION-2.0-14Bncsoft73.4%
- Google: Gemma 3 27Bgoogle70.9%
- Qwen: Qwen2.5 VL 72B Instructqwen70.9%
- rhymes-ai/Ariarhymes-ai69.6%
- openbmb/MiniCPM-V-4openbmb67.1%
- OpenGVLab/InternVL3-8Bopengvlab67.1%
- OpenGVLab/InternVL2_5-4Bopengvlab65.8%
- Qwen/Qwen2.5-VL-7B-Instructqwen65.8%
- Qwen/Qwen2-VL-7B-Instructqwen64.6%
- Xkev/Llama-3.2V-11B-cotxkev64.6%
image_style
mmBench · MMBench · image_style · percent · Higher is better
- OpenGVLab/InternVL2-40Bopengvlab84.7%
- Qwen: Qwen2.5 VL 72B Instructqwen83.5%
- Qwen/Qwen2.5-VL-7B-Instructqwen82.4%
- DeepSeek: DeepSeek-VL2deepseek81.2%
- Google: Gemma 3 12Bgoogle80%
- moonshotai/Kimi-VL-A3B-Instructmoonshotai80%
- Qwen/Qwen2-VL-72B-Instructqwen78.8%
- Qwen/Qwen2-VL-7B-Instructqwen78.8%
- 5CD-AI/Vintern-3B-beta5cd-ai75.3%
- HuggingFaceM4/Idefics3-8B-Llama3huggingfacem475.3%
- OpenGVLab/InternVL3-2Bopengvlab75.3%
- OpenGVLab/InternVL3-8Bopengvlab75.3%
RR
mmBench · MMBench · RR · percent · Higher is better
- OpenGVLab/InternVL2-40Bopengvlab85.7%
- Qwen: Qwen2.5 VL 72B Instructqwen85.3%
- NCSOFT/VARCO-VISION-2.0-14Bncsoft84.5%
- OpenGVLab/InternVL2_5-8Bopengvlab82.9%
- OpenGVLab/InternVL3-8Bopengvlab82.9%
- OpenGVLab/InternVL3-2Bopengvlab81.4%
- Qwen/Qwen2-VL-72B-Instructqwen81.4%
- YannQi/R-4Byannqi81.4%
- Qwen/Qwen2.5-VL-7B-Instructqwen80.2%
- DeepSeek: DeepSeek-VL2deepseek79.8%
- Qwen/Qwen2.5-VL-32B-Instructqwen79.8%
- rhymes-ai/Ariarhymes-ai78.3%
image_topic
mmBench · MMBench · image_topic · percent · Higher is better
- OpenGVLab/InternVL3-8Bopengvlab98.8%
- DeepSeek: DeepSeek-VL2deepseek97.6%
- Google: Gemma 3 12Bgoogle97.6%
- h2oai/h2ovl-mississippi-2bh2oai97.6%
- HuggingFaceM4/Idefics3-8B-Llama3huggingfacem497.6%
- ibm-granite/granite-vision-3.2-2bibm-granite97.6%
- OpenGVLab/InternVL2-40Bopengvlab97.6%
- OpenGVLab/InternVL3-2Bopengvlab97.6%
- rhymes-ai/Ariarhymes-ai97.6%
- Google: Gemma 3 4Bgoogle96.5%
- mistralai/Pixtral-12B-2409mistralai96.5%
- tiiuae/falcon-11B-vlmtiiuae96.5%
physical_relation
mmBench · MMBench · physical_relation · percent · Higher is better
- OpenGVLab/InternVL2-40Bopengvlab72%
- OpenGVLab/InternVL3-8Bopengvlab72%
- DeepSeek: DeepSeek-VL2deepseek70.7%
- NCSOFT/VARCO-VISION-2.0-14Bncsoft70.7%
- OpenGVLab/InternVL2_5-8Bopengvlab70.7%
- YannQi/R-4Byannqi70.7%
- rhymes-ai/Ariarhymes-ai68%
- openbmb/MiniCPM-V-4openbmb66.7%
- Qwen: Qwen2.5 VL 72B Instructqwen66.7%
- moonshotai/Kimi-VL-A3B-Instructmoonshotai65.3%
- OpenGVLab/InternVL2_5-4Bopengvlab64%
- XiaomiMiMo/MiMo-VL-7B-RLxiaomi64%
structuralized_imagetext_understanding
mmBench · MMBench · structuralized_imagetext_understanding · percent · Higher is better
- Qwen/Qwen2.5-VL-32B-Instructqwen88.1%
- YannQi/R-4Byannqi87.2%
- Qwen: Qwen2.5 VL 72B Instructqwen86.2%
- XiaomiMiMo/MiMo-VL-7B-RLxiaomi83.5%
- moonshotai/Kimi-VL-A3B-Thinking-2506moonshotai78.9%
- NCSOFT/VARCO-VISION-2.0-14Bncsoft77.1%
- Qwen/Qwen2-VL-72B-Instructqwen77.1%
- OpenGVLab/InternVL2-40Bopengvlab71.6%
- Qwen/Qwen2.5-VL-7B-Instructqwen67.9%
- moonshotai/Kimi-VL-A3B-Instructmoonshotai67%
- OpenGVLab/InternVL2_5-8Bopengvlab67%
- Qwen/Qwen2.5-VL-3B-Instructqwen67%
physical_property_reasoning
mmBench · MMBench · physical_property_reasoning · percent · Higher is better
- OpenGVLab/InternVL2-40Bopengvlab75%
- Qwen/Qwen2-VL-72B-Instructqwen72%
- DeepSeek: DeepSeek-VL2deepseek71%
- Qwen: Qwen2.5 VL 72B Instructqwen69%
- OpenGVLab/InternVL3-8Bopengvlab67%
- Qwen/Qwen2.5-VL-7B-Instructqwen67%
- Qwen/Qwen2-VL-7B-Instructqwen66%
- OpenGVLab/InternVL3-2Bopengvlab65%
- moonshotai/Kimi-VL-A3B-Instructmoonshotai64%
- nvidia/NVLM-D-72Bnvidia64%
- OpenGVLab/InternVL2-1Bopengvlab64%
- rhymes-ai/Ariarhymes-ai64%
image_scene
mmBench · MMBench · image_scene · percent · Higher is better
- Qwen/Qwen2.5-VL-3B-Instructqwen96.1%
- Qwen: Qwen2.5 VL 72B Instructqwen96.1%
- DeepSeek: DeepSeek-VL2deepseek95.3%
- Qwen/Qwen2.5-VL-7B-Instructqwen95.3%
- Qwen/Qwen2-VL-72B-Instructqwen95.3%
- moonshotai/Kimi-VL-A3B-Instructmoonshotai94.6%
- OpenGVLab/InternVL2-40Bopengvlab94.6%
- 5CD-AI/Vintern-3B-beta5cd-ai93.8%
- Google: Gemma 3 4Bgoogle93.8%
- Qwen/Qwen2-VL-2B-Instructqwen93.8%
- HuggingFaceM4/Idefics3-8B-Llama3huggingfacem493%
- tiiuae/falcon-11B-vlmtiiuae93%
action_recognition
mmBench · MMBench · action_recognition · percent · Higher is better
- OpenGVLab/InternVL2_5-8Bopengvlab93.2%
- OpenGVLab/InternVL3-8Bopengvlab93.2%
- OpenGVLab/InternVL2-40Bopengvlab92.3%
- Qwen/Qwen2-VL-72B-Instructqwen92.3%
- nvidia/NVLM-D-72Bnvidia90.6%
- OpenGVLab/InternVL3-2Bopengvlab90.6%
- Qwen/Qwen2-VL-7B-Instructqwen89.7%
- YannQi/R-4Byannqi89.7%
- Llama-3.2-90B-Vision-Instructmeta-llama88.9%
- HuggingFaceM4/idefics2-8bhuggingfacem488%
- ibm-granite/granite-vision-3.2-2bibm-granite88%
- Qwen/Qwen2-VL-2B-Instructqwen88%
food_clothes
mmBench · MMBench · food_clothes · percent · Higher is better
- OpenGVLab/InternVL2-40Bopengvlab80%
- OpenGVLab/InternVL2-1Bopengvlab73.9%
- OpenGVLab/InternVL3-8Bopengvlab69.6%
- OpenGVLab/InternVL3-2Bopengvlab67.8%
- moonshotai/Kimi-VL-A3B-Instructmoonshotai65.2%
- Qwen/Qwen2-VL-7B-Instructqwen63.5%
- Qwen: Qwen2.5 VL 72B Instructqwen60%
- Qwen/Qwen2-VL-72B-Instructqwen57.4%
- nvidia/NVLM-D-72Bnvidia55.7%
- Qwen/Qwen2.5-VL-7B-Instructqwen54.8%
- Qwen/Qwen2.5-VL-3B-Instructqwen53%
- Qwen/Qwen2-VL-2B-Instructqwen53%
Models
LoadingBenchmarks
Explore every benchmark
LoadingThese charts contain model-linked results. The charts above also include unlinked source profiles. Compare scores within each evaluation. Model quality, usage, transparency and system performance measure different things.