MMBench

Indexed results from MMBench. Each ranking retains its own metric and evaluation conditions.

nature_relation

mmBench · MMBench · nature_relation · percent · Higher is better

  1. OpenGVLab/InternVL2-40Bopengvlab93.5%
  2. Qwen: Qwen2.5 VL 72B Instructqwen93.5%
  3. NCSOFT/VARCO-VISION-2.0-14Bncsoft91.3%
  4. YannQi/R-4Byannqi91.3%
  5. Google: Gemma 3 27Bgoogle90.2%
  6. Qwen/Qwen2.5-VL-7B-Instructqwen89.1%
  7. OpenGVLab/InternVL2_5-8Bopengvlab88%
  8. OpenGVLab/InternVL3-8Bopengvlab88%
  9. Qwen/Qwen2-VL-72B-Instructqwen88%
  10. DeepSeek: DeepSeek-VL2deepseek85.9%
  11. OpenGVLab/InternVL3-2Bopengvlab85.9%
  12. Qwen/Qwen2.5-VL-32B-Instructqwen85.9%

image_emotion

mmBench · MMBench · image_emotion · percent · Higher is better

  1. OpenGVLab/InternVL2-40Bopengvlab84.4%
  2. OpenGVLab/InternVL2_5-4Bopengvlab83.3%
  3. Qwen/Qwen2-VL-72B-Instructqwen83.3%
  4. Qwen/Qwen2-VL-7B-Instructqwen83.3%
  5. OpenGVLab/InternVL2_5-8Bopengvlab82.2%
  6. OpenGVLab/InternVL3-8Bopengvlab82.2%
  7. Qwen: Qwen2.5 VL 72B Instructqwen82.2%
  8. ibm-granite/granite-vision-3.2-2bibm-granite81.1%
  9. Google: Gemma 3 27Bgoogle80%
  10. HuggingFaceM4/Idefics3-8B-Llama3huggingfacem480%
  11. ibm-granite/granite-vision-3.1-2b-previewibm-granite80%
  12. NCSOFT/VARCO-VISION-2.0-14Bncsoft80%

physical_property_reasoning

mmBench · MMBench · physical_property_reasoning · percent · Higher is better

  1. OpenGVLab/InternVL2-40Bopengvlab79.7%
  2. Qwen/Qwen2-VL-72B-Instructqwen77.2%
  3. NCSOFT/VARCO-VISION-2.0-14Bncsoft73.4%
  4. Google: Gemma 3 27Bgoogle70.9%
  5. Qwen: Qwen2.5 VL 72B Instructqwen70.9%
  6. rhymes-ai/Ariarhymes-ai69.6%
  7. openbmb/MiniCPM-V-4openbmb67.1%
  8. OpenGVLab/InternVL3-8Bopengvlab67.1%
  9. OpenGVLab/InternVL2_5-4Bopengvlab65.8%
  10. Qwen/Qwen2.5-VL-7B-Instructqwen65.8%
  11. Qwen/Qwen2-VL-7B-Instructqwen64.6%
  12. Xkev/Llama-3.2V-11B-cotxkev64.6%

image_style

mmBench · MMBench · image_style · percent · Higher is better

  1. OpenGVLab/InternVL2-40Bopengvlab84.7%
  2. Qwen: Qwen2.5 VL 72B Instructqwen83.5%
  3. Qwen/Qwen2.5-VL-7B-Instructqwen82.4%
  4. DeepSeek: DeepSeek-VL2deepseek81.2%
  5. Google: Gemma 3 12Bgoogle80%
  6. moonshotai/Kimi-VL-A3B-Instructmoonshotai80%
  7. Qwen/Qwen2-VL-72B-Instructqwen78.8%
  8. Qwen/Qwen2-VL-7B-Instructqwen78.8%
  9. 5CD-AI/Vintern-3B-beta5cd-ai75.3%
  10. HuggingFaceM4/Idefics3-8B-Llama3huggingfacem475.3%
  11. OpenGVLab/InternVL3-2Bopengvlab75.3%
  12. OpenGVLab/InternVL3-8Bopengvlab75.3%

RR

mmBench · MMBench · RR · percent · Higher is better

  1. OpenGVLab/InternVL2-40Bopengvlab85.7%
  2. Qwen: Qwen2.5 VL 72B Instructqwen85.3%
  3. NCSOFT/VARCO-VISION-2.0-14Bncsoft84.5%
  4. OpenGVLab/InternVL2_5-8Bopengvlab82.9%
  5. OpenGVLab/InternVL3-8Bopengvlab82.9%
  6. OpenGVLab/InternVL3-2Bopengvlab81.4%
  7. Qwen/Qwen2-VL-72B-Instructqwen81.4%
  8. YannQi/R-4Byannqi81.4%
  9. Qwen/Qwen2.5-VL-7B-Instructqwen80.2%
  10. DeepSeek: DeepSeek-VL2deepseek79.8%
  11. Qwen/Qwen2.5-VL-32B-Instructqwen79.8%
  12. rhymes-ai/Ariarhymes-ai78.3%

image_topic

mmBench · MMBench · image_topic · percent · Higher is better

  1. OpenGVLab/InternVL3-8Bopengvlab98.8%
  2. DeepSeek: DeepSeek-VL2deepseek97.6%
  3. Google: Gemma 3 12Bgoogle97.6%
  4. h2oai/h2ovl-mississippi-2bh2oai97.6%
  5. HuggingFaceM4/Idefics3-8B-Llama3huggingfacem497.6%
  6. ibm-granite/granite-vision-3.2-2bibm-granite97.6%
  7. OpenGVLab/InternVL2-40Bopengvlab97.6%
  8. OpenGVLab/InternVL3-2Bopengvlab97.6%
  9. rhymes-ai/Ariarhymes-ai97.6%
  10. Google: Gemma 3 4Bgoogle96.5%
  11. mistralai/Pixtral-12B-2409mistralai96.5%
  12. tiiuae/falcon-11B-vlmtiiuae96.5%

physical_relation

mmBench · MMBench · physical_relation · percent · Higher is better

  1. OpenGVLab/InternVL2-40Bopengvlab72%
  2. OpenGVLab/InternVL3-8Bopengvlab72%
  3. DeepSeek: DeepSeek-VL2deepseek70.7%
  4. NCSOFT/VARCO-VISION-2.0-14Bncsoft70.7%
  5. OpenGVLab/InternVL2_5-8Bopengvlab70.7%
  6. YannQi/R-4Byannqi70.7%
  7. rhymes-ai/Ariarhymes-ai68%
  8. openbmb/MiniCPM-V-4openbmb66.7%
  9. Qwen: Qwen2.5 VL 72B Instructqwen66.7%
  10. moonshotai/Kimi-VL-A3B-Instructmoonshotai65.3%
  11. OpenGVLab/InternVL2_5-4Bopengvlab64%
  12. XiaomiMiMo/MiMo-VL-7B-RLxiaomi64%

structuralized_imagetext_understanding

mmBench · MMBench · structuralized_imagetext_understanding · percent · Higher is better

  1. Qwen/Qwen2.5-VL-32B-Instructqwen88.1%
  2. YannQi/R-4Byannqi87.2%
  3. Qwen: Qwen2.5 VL 72B Instructqwen86.2%
  4. XiaomiMiMo/MiMo-VL-7B-RLxiaomi83.5%
  5. moonshotai/Kimi-VL-A3B-Thinking-2506moonshotai78.9%
  6. NCSOFT/VARCO-VISION-2.0-14Bncsoft77.1%
  7. Qwen/Qwen2-VL-72B-Instructqwen77.1%
  8. OpenGVLab/InternVL2-40Bopengvlab71.6%
  9. Qwen/Qwen2.5-VL-7B-Instructqwen67.9%
  10. moonshotai/Kimi-VL-A3B-Instructmoonshotai67%
  11. OpenGVLab/InternVL2_5-8Bopengvlab67%
  12. Qwen/Qwen2.5-VL-3B-Instructqwen67%

physical_property_reasoning

mmBench · MMBench · physical_property_reasoning · percent · Higher is better

  1. OpenGVLab/InternVL2-40Bopengvlab75%
  2. Qwen/Qwen2-VL-72B-Instructqwen72%
  3. DeepSeek: DeepSeek-VL2deepseek71%
  4. Qwen: Qwen2.5 VL 72B Instructqwen69%
  5. OpenGVLab/InternVL3-8Bopengvlab67%
  6. Qwen/Qwen2.5-VL-7B-Instructqwen67%
  7. Qwen/Qwen2-VL-7B-Instructqwen66%
  8. OpenGVLab/InternVL3-2Bopengvlab65%
  9. moonshotai/Kimi-VL-A3B-Instructmoonshotai64%
  10. nvidia/NVLM-D-72Bnvidia64%
  11. OpenGVLab/InternVL2-1Bopengvlab64%
  12. rhymes-ai/Ariarhymes-ai64%

image_scene

mmBench · MMBench · image_scene · percent · Higher is better

  1. Qwen/Qwen2.5-VL-3B-Instructqwen96.1%
  2. Qwen: Qwen2.5 VL 72B Instructqwen96.1%
  3. DeepSeek: DeepSeek-VL2deepseek95.3%
  4. Qwen/Qwen2.5-VL-7B-Instructqwen95.3%
  5. Qwen/Qwen2-VL-72B-Instructqwen95.3%
  6. moonshotai/Kimi-VL-A3B-Instructmoonshotai94.6%
  7. OpenGVLab/InternVL2-40Bopengvlab94.6%
  8. 5CD-AI/Vintern-3B-beta5cd-ai93.8%
  9. Google: Gemma 3 4Bgoogle93.8%
  10. Qwen/Qwen2-VL-2B-Instructqwen93.8%
  11. HuggingFaceM4/Idefics3-8B-Llama3huggingfacem493%
  12. tiiuae/falcon-11B-vlmtiiuae93%

action_recognition

mmBench · MMBench · action_recognition · percent · Higher is better

  1. OpenGVLab/InternVL2_5-8Bopengvlab93.2%
  2. OpenGVLab/InternVL3-8Bopengvlab93.2%
  3. OpenGVLab/InternVL2-40Bopengvlab92.3%
  4. Qwen/Qwen2-VL-72B-Instructqwen92.3%
  5. nvidia/NVLM-D-72Bnvidia90.6%
  6. OpenGVLab/InternVL3-2Bopengvlab90.6%
  7. Qwen/Qwen2-VL-7B-Instructqwen89.7%
  8. YannQi/R-4Byannqi89.7%
  9. Llama-3.2-90B-Vision-Instructmeta-llama88.9%
  10. HuggingFaceM4/idefics2-8bhuggingfacem488%
  11. ibm-granite/granite-vision-3.2-2bibm-granite88%
  12. Qwen/Qwen2-VL-2B-Instructqwen88%

food_clothes

mmBench · MMBench · food_clothes · percent · Higher is better

  1. OpenGVLab/InternVL2-40Bopengvlab80%
  2. OpenGVLab/InternVL2-1Bopengvlab73.9%
  3. OpenGVLab/InternVL3-8Bopengvlab69.6%
  4. OpenGVLab/InternVL3-2Bopengvlab67.8%
  5. moonshotai/Kimi-VL-A3B-Instructmoonshotai65.2%
  6. Qwen/Qwen2-VL-7B-Instructqwen63.5%
  7. Qwen: Qwen2.5 VL 72B Instructqwen60%
  8. Qwen/Qwen2-VL-72B-Instructqwen57.4%
  9. nvidia/NVLM-D-72Bnvidia55.7%
  10. Qwen/Qwen2.5-VL-7B-Instructqwen54.8%
  11. Qwen/Qwen2.5-VL-3B-Instructqwen53%
  12. Qwen/Qwen2-VL-2B-Instructqwen53%

Models

Loading
EVIDENCE / COMPARISON

Benchmarks

Explore every benchmark

Loading

These 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.

Install TensorTrace