MMBench

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

function_reasoning

mmBench · MMBench · function_reasoning · percent · Higher is better

  1. Llama-3.2-90B-Vision-Instructmeta-llama97.8%
  2. DeepSeek: DeepSeek-VL2deepseek96.7%
  3. OpenGVLab/InternVL2-40Bopengvlab96.7%
  4. Google: Gemma 3 12Bgoogle95.6%
  5. moonshotai/Kimi-VL-A3B-Instructmoonshotai95.6%
  6. Qwen/Qwen2.5-VL-32B-Instructqwen95.6%
  7. Qwen/Qwen2.5-VL-3B-Instructqwen95.6%
  8. Qwen: Qwen2.5 VL 72B Instructqwen95.6%
  9. Qwen/Qwen2.5-VL-7B-Instructqwen95.6%
  10. mistralai/Pixtral-12B-2409mistralai94.4%
  11. moonshotai/Kimi-VL-A3B-Thinking-2506moonshotai94.4%
  12. Qwen/Qwen2-VL-72B-Instructqwen94.4%

image_emotion

mmBench · MMBench · image_emotion · percent · Higher is better

  1. OpenGVLab/InternVL2-40Bopengvlab87.8%
  2. Qwen/Qwen2-VL-7B-Instructqwen86.7%
  3. openbmb/MiniCPM-V-4openbmb85.6%
  4. OpenGVLab/InternVL2_5-4Bopengvlab85.6%
  5. OpenGVLab/InternVL3-8Bopengvlab85.6%
  6. OpenGVLab/InternVL2_5-8Bopengvlab84.4%
  7. HuggingFaceM4/Idefics3-8B-Llama3huggingfacem483.3%
  8. Qwen/Qwen2-VL-72B-Instructqwen82.2%
  9. tiiuae/falcon-11B-vlmtiiuae82.2%
  10. DeepSeek: DeepSeek-VL2deepseek81.1%
  11. Google: Gemma 3 27Bgoogle81.1%
  12. Qwen: Qwen2.5 VL 72B Instructqwen81.1%

FP-C

mmBench · MMBench · FP-C · percent · Higher is better

  1. Qwen/Qwen2-VL-72B-Instructqwen88%
  2. Qwen: Qwen2.5 VL 72B Instructqwen85.8%
  3. YannQi/R-4Byannqi82.4%
  4. NCSOFT/VARCO-VISION-2.0-14Bncsoft81.6%
  5. OpenGVLab/InternVL2-40Bopengvlab81.3%
  6. OpenGVLab/InternVL3-8Bopengvlab79%
  7. Qwen/Qwen2.5-VL-32B-Instructqwen79%
  8. DeepSeek: DeepSeek-VL2deepseek78.3%
  9. OpenGVLab/InternVL2_5-8Bopengvlab78.3%
  10. Qwen/Qwen2.5-VL-7B-Instructqwen77.2%
  11. nvidia/NVLM-D-72Bnvidia76.4%
  12. Qwen/Qwen2-VL-7B-Instructqwen76.4%

future_prediction

mmBench · MMBench · future_prediction · percent · Higher is better

  1. OpenGVLab/InternVL2-40Bopengvlab80.6%
  2. Qwen/Qwen2-VL-72B-Instructqwen80.6%
  3. Qwen: Qwen2.5 VL 72B Instructqwen77.8%
  4. DeepSeek: DeepSeek-VL2deepseek75%
  5. OpenGVLab/InternVL3-8Bopengvlab75%
  6. Google: Gemma 3 12Bgoogle73.6%
  7. moonshotai/Kimi-VL-A3B-Instructmoonshotai73.6%
  8. Qwen/Qwen2.5-VL-3B-Instructqwen73.6%
  9. Qwen/Qwen2.5-VL-7B-Instructqwen73.6%
  10. OpenGVLab/InternVL3-2Bopengvlab72.2%
  11. rhymes-ai/Ariarhymes-ai70.8%
  12. Qwen/Qwen2-VL-7B-Instructqwen69.4%

image_scene

mmBench · MMBench · image_scene · percent · Higher is better

  1. Qwen: Qwen2.5 VL 72B Instructqwen93.8%
  2. Qwen/Qwen2.5-VL-32B-Instructqwen92.4%
  3. OpenGVLab/InternVL2_5-8Bopengvlab91.7%
  4. Qwen/Qwen2.5-VL-3B-Instructqwen91%
  5. NCSOFT/VARCO-VISION-2.0-14Bncsoft90.3%
  6. Qwen/Qwen2.5-VL-7B-Instructqwen90.3%
  7. tiiuae/falcon-11B-vlmtiiuae89.6%
  8. vikhyatk/moondream2vikhyatk89.6%
  9. XiaomiMiMo/MiMo-VL-7B-RLxiaomi89.6%
  10. microsoft/Phi-3.5-vision-instructmicrosoft88.9%
  11. microsoft/Phi-3-vision-128k-instructmicrosoft88.9%
  12. moonshotai/Kimi-VL-A3B-Instructmoonshotai88.9%

object_localization

mmBench · MMBench · object_localization · percent · Higher is better

  1. Qwen: Qwen2.5 VL 72B Instructqwen90.4%
  2. OpenGVLab/InternVL2-40Bopengvlab86.5%
  3. Qwen/Qwen2-VL-72B-Instructqwen84.6%
  4. Llama-3.2-90B-Vision-Instructmeta-llama82.7%
  5. moonshotai/Kimi-VL-A3B-Instructmoonshotai80.8%
  6. OpenGVLab/InternVL3-8Bopengvlab79.8%
  7. Qwen/Qwen2.5-VL-7B-Instructqwen78.8%
  8. nvidia/NVLM-D-72Bnvidia76.9%
  9. rhymes-ai/Ariarhymes-ai76%
  10. DeepSeek: DeepSeek-VL2deepseek74%
  11. OpenGVLab/InternVL3-2Bopengvlab73.1%
  12. ibm-granite/granite-vision-3.2-2bibm-granite72.1%

attribute_comparison

mmBench · MMBench · attribute_comparison · percent · Higher is better

  1. NCSOFT/VARCO-VISION-2.0-14Bncsoft90.7%
  2. Qwen: Qwen2.5 VL 72B Instructqwen90.7%
  3. XiaomiMiMo/MiMo-VL-7B-RLxiaomi90.7%
  4. YannQi/R-4Byannqi90.7%
  5. Qwen/Qwen2.5-VL-32B-Instructqwen89.3%
  6. Qwen/Qwen2-VL-72B-Instructqwen89.3%
  7. openbmb/MiniCPM-V-4openbmb88%
  8. OpenGVLab/InternVL2_5-8Bopengvlab88%
  9. DeepSeek: DeepSeek-VL2deepseek86.7%
  10. moonshotai/Kimi-VL-A3B-Instructmoonshotai84%
  11. Qwen/Qwen2.5-VL-7B-Instructqwen84%
  12. moonshotai/Kimi-VL-A3B-Thinking-2506moonshotai82.7%

object_localization

mmBench · MMBench · object_localization · percent · Higher is better

  1. Qwen: Qwen2.5 VL 72B Instructqwen91.3%
  2. Qwen/Qwen2-VL-72B-Instructqwen86.5%
  3. OpenGVLab/InternVL2-40Bopengvlab81.7%
  4. Qwen/Qwen2.5-VL-7B-Instructqwen81.7%
  5. moonshotai/Kimi-VL-A3B-Instructmoonshotai80.8%
  6. OpenGVLab/InternVL3-8Bopengvlab77.9%
  7. DeepSeek: DeepSeek-VL2deepseek76.9%
  8. OpenGVLab/InternVL3-2Bopengvlab75%
  9. rhymes-ai/Ariarhymes-ai74%
  10. Llama-3.2-90B-Vision-Instructmeta-llama73.1%
  11. nvidia/NVLM-D-72Bnvidia72.1%
  12. Qwen/Qwen2-VL-7B-Instructqwen67.3%

FP-S

mmBench · MMBench · FP-S · percent · Higher is better

  1. Qwen: Qwen2.5 VL 72B Instructqwen95.1%
  2. OpenGVLab/InternVL2-40Bopengvlab93.3%
  3. Qwen/Qwen2-VL-72B-Instructqwen92.3%
  4. Qwen/Qwen2.5-VL-32B-Instructqwen91%
  5. NCSOFT/VARCO-VISION-2.0-14Bncsoft90.3%
  6. OpenGVLab/InternVL2_5-8Bopengvlab90%
  7. Qwen/Qwen2.5-VL-7B-Instructqwen89.8%
  8. moonshotai/Kimi-VL-A3B-Instructmoonshotai89.3%
  9. Qwen/Qwen2-VL-7B-Instructqwen89.3%
  10. YannQi/R-4Byannqi88.9%
  11. XiaomiMiMo/MiMo-VL-7B-RLxiaomi88.6%
  12. DeepSeek: DeepSeek-VL2deepseek88.4%

AR

mmBench · MMBench · AR · percent · Higher is better

  1. Qwen/Qwen2-VL-72B-Instructqwen89.4%
  2. Qwen/Qwen2.5-VL-32B-Instructqwen88.6%
  3. rhymes-ai/Ariarhymes-ai88.6%
  4. Qwen: Qwen2.5 VL 72B Instructqwen88.2%
  5. DeepSeek: DeepSeek-VL2deepseek87.3%
  6. moonshotai/Kimi-VL-A3B-Instructmoonshotai87.3%
  7. NCSOFT/VARCO-VISION-2.0-14Bncsoft87.3%
  8. OpenGVLab/InternVL2-40Bopengvlab87.3%
  9. Google: Gemma 3 27Bgoogle86.9%
  10. Llama-3.2-90B-Vision-Instructmeta-llama86.5%
  11. openbmb/MiniCPM-V-4openbmb85.7%
  12. YannQi/R-4Byannqi85.7%

image_quality

mmBench · MMBench · image_quality · percent · Higher is better

  1. Qwen: Qwen2.5 VL 72B Instructqwen88.2%
  2. DeepSeek: DeepSeek-VL2deepseek85.9%
  3. OpenGVLab/InternVL2-40Bopengvlab84.7%
  4. Qwen/Qwen2-VL-72B-Instructqwen83.5%
  5. Qwen/Qwen2.5-VL-7B-Instructqwen82.4%
  6. OpenGVLab/InternVL3-8Bopengvlab81.2%
  7. moonshotai/Kimi-VL-A3B-Instructmoonshotai80%
  8. nvidia/NVLM-D-72Bnvidia78.8%
  9. rhymes-ai/Ariarhymes-ai78.8%
  10. ibm-granite/granite-vision-3.2-2bibm-granite76.5%
  11. Llama-3.2-90B-Vision-Instructmeta-llama76.5%
  12. OpenGVLab/InternVL3-2Bopengvlab76.5%

identity_reasoning

mmBench · MMBench · identity_reasoning · percent · Higher is better

  1. Google: Gemma 3 27Bgoogle100%
  2. moonshotai/Kimi-VL-A3B-Thinking-2506moonshotai98.7%
  3. NCSOFT/VARCO-VISION-2.0-14Bncsoft98.7%
  4. nvidia/NVLM-D-72Bnvidia98.7%
  5. OpenGVLab/InternVL2_5-8Bopengvlab98.7%
  6. OpenGVLab/InternVL3-8Bopengvlab98.7%
  7. Qwen/Qwen2.5-VL-32B-Instructqwen98.7%
  8. Qwen/Qwen2-VL-7B-Instructqwen98.7%
  9. Google: Gemma 3 12Bgoogle97.4%
  10. HuggingFaceM4/Idefics3-8B-Llama3huggingfacem497.4%
  11. Qwen/Qwen2.5-VL-7B-Instructqwen97.4%
  12. Qwen/Qwen2-VL-2B-Instructqwen97.4%

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