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. NCSOFT/VARCO-VISION-2.0-14Bncsoft93.5%
  2. OpenGVLab/InternVL2-40Bopengvlab92.4%
  3. Qwen: Qwen2.5 VL 72B Instructqwen91.3%
  4. YannQi/R-4Byannqi90.2%
  5. Google: Gemma 3 12Bgoogle89.1%
  6. Google: Gemma 3 27Bgoogle89.1%
  7. OpenGVLab/InternVL3-8Bopengvlab89.1%
  8. Qwen/Qwen2-VL-72B-Instructqwen89.1%
  9. Llama-3.2-90B-Vision-Instructmeta-llama88%
  10. moonshotai/Kimi-VL-A3B-Thinking-2506moonshotai88%
  11. Qwen/Qwen2.5-VL-32B-Instructqwen88%
  12. OpenGVLab/InternVL2_5-8Bopengvlab87%

object_localization

mmBench · MMBench · object_localization · percent · Higher is better

  1. Qwen/Qwen2.5-VL-32B-Instructqwen90.5%
  2. Qwen: Qwen2.5 VL 72B Instructqwen89.5%
  3. NCSOFT/VARCO-VISION-2.0-14Bncsoft88.6%
  4. OpenGVLab/InternVL2-40Bopengvlab88.6%
  5. OpenGVLab/InternVL2_5-8Bopengvlab85.7%
  6. Qwen/Qwen2-VL-72B-Instructqwen85.7%
  7. moonshotai/Kimi-VL-A3B-Instructmoonshotai83.8%
  8. XiaomiMiMo/MiMo-VL-7B-RLxiaomi82.9%
  9. YannQi/R-4Byannqi82.9%
  10. OpenGVLab/InternVL3-8Bopengvlab81.9%
  11. Qwen/Qwen2.5-VL-7B-Instructqwen81.9%
  12. nvidia/NVLM-D-72Bnvidia80%

action_recognition

mmBench · MMBench · action_recognition · percent · Higher is better

  1. OpenGVLab/InternVL3-8Bopengvlab96.6%
  2. Qwen/Qwen2-VL-72B-Instructqwen95.5%
  3. DeepSeek: DeepSeek-VL2deepseek94.3%
  4. h2oai/h2ovl-mississippi-2bh2oai94.3%
  5. OpenGVLab/InternVL2-1Bopengvlab94.3%
  6. OpenGVLab/InternVL3-2Bopengvlab94.3%
  7. Qwen/Qwen2.5-VL-7B-Instructqwen94.3%
  8. Qwen/Qwen2-VL-2B-Instructqwen94.3%
  9. Qwen/Qwen2-VL-7B-Instructqwen94.3%
  10. tiiuae/falcon-11B-vlmtiiuae94.3%
  11. moonshotai/Kimi-VL-A3B-Instructmoonshotai93.2%
  12. nvidia/NVLM-D-72Bnvidia93.2%

image_scene

mmBench · MMBench · image_scene · percent · Higher is better

  1. OpenGVLab/InternVL2_5-8Bopengvlab92.4%
  2. Qwen/Qwen2.5-VL-3B-Instructqwen92.4%
  3. Qwen/Qwen2.5-VL-7B-Instructqwen92.4%
  4. OpenGVLab/InternVL2_5-4Bopengvlab91.7%
  5. Qwen: Qwen2.5 VL 72B Instructqwen91.7%
  6. NCSOFT/VARCO-VISION-2.0-14Bncsoft90.3%
  7. openbmb/MiniCPM-V-4openbmb90.3%
  8. DeepSeek: DeepSeek-VL2deepseek89.6%
  9. moonshotai/Kimi-VL-A3B-Instructmoonshotai89.6%
  10. OpenGVLab/InternVL2-40Bopengvlab89.6%
  11. Qwen/Qwen2.5-VL-32B-Instructqwen88.9%
  12. tiiuae/falcon-11B-vlmtiiuae88.9%

nature_relation

mmBench · MMBench · nature_relation · percent · Higher is better

  1. OpenGVLab/InternVL2-40Bopengvlab95.1%
  2. Qwen: Qwen2.5 VL 72B Instructqwen91.4%
  3. OpenGVLab/InternVL3-2Bopengvlab87.7%
  4. Qwen/Qwen2.5-VL-7B-Instructqwen87.7%
  5. Qwen/Qwen2-VL-72B-Instructqwen86.4%
  6. DeepSeek: DeepSeek-VL2deepseek85.2%
  7. OpenGVLab/InternVL3-8Bopengvlab85.2%
  8. rhymes-ai/Ariarhymes-ai84%
  9. nvidia/NVLM-D-72Bnvidia82.7%
  10. Google: Gemma 3 12Bgoogle81.5%
  11. Llama-3.2-90B-Vision-Instructmeta-llama79%
  12. microsoft/Phi-3.5-vision-instructmicrosoft79%

image_topic

mmBench · MMBench · image_topic · percent · Higher is better

  1. HuggingFaceM4/idefics2-8bhuggingfacem498.9%
  2. mistralai/Pixtral-12B-2409mistralai98.9%
  3. Qwen/Qwen2.5-VL-32B-Instructqwen98.9%
  4. Qwen: Qwen2.5 VL 72B Instructqwen98.9%
  5. Qwen/Qwen2-VL-2B-Instructqwen98.9%
  6. rhymes-ai/Ariarhymes-ai98.9%
  7. Google: Gemma 3 12Bgoogle97.8%
  8. Google: Gemma 3 27Bgoogle97.8%
  9. ibm-granite/granite-vision-3.2-2bibm-granite97.8%
  10. Qwen/Qwen2.5-VL-3B-Instructqwen97.8%
  11. Qwen/Qwen2.5-VL-7B-Instructqwen97.8%
  12. tiiuae/falcon-11B-vlmtiiuae97.8%

ocr

mmBench · MMBench · ocr · percent · Higher is better

  1. Qwen: Qwen2.5 VL 72B Instructqwen100%
  2. nvidia/NVLM-D-72Bnvidia98.7%
  3. moonshotai/Kimi-VL-A3B-Instructmoonshotai97.4%
  4. Qwen/Qwen2-VL-72B-Instructqwen97.4%
  5. Llama-3.2-90B-Vision-Instructmeta-llama96.1%
  6. OpenGVLab/InternVL2-40Bopengvlab96.1%
  7. OpenGVLab/InternVL3-8Bopengvlab96.1%
  8. Qwen/Qwen2.5-VL-7B-Instructqwen96.1%
  9. Qwen/Qwen2-VL-7B-Instructqwen96.1%
  10. rhymes-ai/Ariarhymes-ai96.1%
  11. Google: Gemma 3 4Bgoogle94.8%
  12. Qwen/Qwen2.5-VL-3B-Instructqwen94.8%

image_style

mmBench · MMBench · image_style · percent · Higher is better

  1. Qwen: Qwen2.5 VL 72B Instructqwen83.7%
  2. Qwen/Qwen2.5-VL-7B-Instructqwen82.6%
  3. Qwen/Qwen2-VL-7B-Instructqwen82.6%
  4. Google: Gemma 3 27Bgoogle81.5%
  5. OpenGVLab/InternVL2-40Bopengvlab81.5%
  6. OpenGVLab/InternVL2_5-8Bopengvlab81.5%
  7. Qwen/Qwen2-VL-72B-Instructqwen80.4%
  8. Google: Gemma 3 12Bgoogle79.3%
  9. moonshotai/Kimi-VL-A3B-Instructmoonshotai79.3%
  10. DeepSeek: DeepSeek-VL2deepseek78.3%
  11. OpenGVLab/InternVL2_5-4Bopengvlab78.3%
  12. Qwen/Qwen2.5-VL-32B-Instructqwen78.3%

physical_property_reasoning

mmBench · MMBench · physical_property_reasoning · percent · Higher is better

  1. DeepSeek: DeepSeek-VL2deepseek76%
  2. OpenGVLab/InternVL2-40Bopengvlab73%
  3. rhymes-ai/Ariarhymes-ai73%
  4. HuggingFaceM4/Idefics3-8B-Llama3huggingfacem471%
  5. Qwen/Qwen2-VL-72B-Instructqwen70%
  6. moonshotai/Kimi-VL-A3B-Instructmoonshotai69%
  7. OpenGVLab/InternVL2-1Bopengvlab68%
  8. OpenGVLab/InternVL3-8Bopengvlab68%
  9. Qwen: Qwen2.5 VL 72B Instructqwen66%
  10. 5CD-AI/Vintern-3B-beta5cd-ai65%
  11. Qwen/Qwen2.5-VL-7B-Instructqwen65%
  12. tiiuae/falcon-11B-vlmtiiuae65%

spatial_relationship

mmBench · MMBench · spatial_relationship · percent · Higher is better

  1. Qwen/Qwen2-VL-72B-Instructqwen81.5%
  2. Qwen: Qwen2.5 VL 72B Instructqwen80.2%
  3. nvidia/NVLM-D-72Bnvidia65.4%
  4. OpenGVLab/InternVL2-40Bopengvlab65.4%
  5. Qwen/Qwen2.5-VL-7B-Instructqwen61.7%
  6. OpenGVLab/InternVL3-8Bopengvlab56.8%
  7. Llama-3.2-90B-Vision-Instructmeta-llama55.6%
  8. DeepSeek: DeepSeek-VL2deepseek49.4%
  9. mistralai/Pixtral-12B-2409mistralai48.1%
  10. rhymes-ai/Ariarhymes-ai48.1%
  11. Google: Gemma 3 12Bgoogle46.9%
  12. HuggingFaceM4/Idefics3-8B-Llama3huggingfacem442%

AR

mmBench · MMBench · AR · percent · Higher is better

  1. DeepSeek: DeepSeek-VL2deepseek87.5%
  2. OpenGVLab/InternVL2-40Bopengvlab86.5%
  3. Qwen/Qwen2-VL-72B-Instructqwen85.4%
  4. rhymes-ai/Ariarhymes-ai85.4%
  5. moonshotai/Kimi-VL-A3B-Instructmoonshotai85.1%
  6. Qwen: Qwen2.5 VL 72B Instructqwen84%
  7. HuggingFaceM4/Idefics3-8B-Llama3huggingfacem483.7%
  8. OpenGVLab/InternVL3-8Bopengvlab83.3%
  9. Llama-3.2-90B-Vision-Instructmeta-llama82.6%
  10. OpenGVLab/InternVL3-2Bopengvlab82.6%
  11. Qwen/Qwen2.5-VL-3B-Instructqwen82.6%
  12. Qwen/Qwen2.5-VL-7B-Instructqwen82.6%

future_prediction

mmBench · MMBench · future_prediction · percent · Higher is better

  1. Qwen/Qwen2-VL-72B-Instructqwen79.2%
  2. OpenGVLab/InternVL2-40Bopengvlab77.8%
  3. Qwen: Qwen2.5 VL 72B Instructqwen77.8%
  4. nvidia/NVLM-D-72Bnvidia76.4%
  5. Qwen/Qwen2.5-VL-7B-Instructqwen76.4%
  6. DeepSeek: DeepSeek-VL2deepseek75%
  7. Google: Gemma 3 12Bgoogle75%
  8. OpenGVLab/InternVL3-8Bopengvlab75%
  9. Google: Gemma 3 4Bgoogle73.6%
  10. Llama-3.2-90B-Vision-Instructmeta-llama73.6%
  11. rhymes-ai/Ariarhymes-ai73.6%
  12. vikhyatk/moondream2vikhyatk72.2%

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