Fine-Grained Perception on MMVP
74.9AccuracyVisual Para-Thinker++
Evaluation results
| Method | Links | |
|---|---|---|
| Visual Para-Thinker++Size=7B2026.06 | 74.9 | |
| UniMRGType=Unified, # LLM Params=2B2026.01 | 74.67 | |
| ReGuLaR-7BModel Category=Latent reasoning LVLMs2026.05 | 73.67 | |
| LLaVA-OneVisionModel Category=General-purpose LVLMs2026.05 | 73.33 | |
| UniLIPType=Unified, # LLM Params=2B2026.01 | 73 | |
| LaserModel Category=Latent reasoning LVLMs2026.05 | 73 | |
| MonetModel Category=Latent reasoning LVLMs2026.05 | 72.33 | |
| Qwen2.5-VLSize=72B2026.06 | 72.1 | |
| Qwen3-VLType=Und. Only, # LLM Params=2B2026.01 | 72 | |
| Vision-R1Model Category=RL-based reasoning LVLMs2026.05 | 72 | |
| InternVL3Type=Und. Only, # LLM Params=2B2026.01 | 71.67 | |
| OpenUniType=Unified, # LLM Params=2B2026.01 | 71.67 | |
| DeepeyesModel Category=RL-based reasoning LVLMs2026.05 | 71.67 | |
| Visual Para-ThinkerSize=7B2026.06 | 71.3 | |
| GRPOSize=7B2026.06 | 70.9 | |
| Qwen2.5-VL-7BModel Category=General-purpose LVLMs2026.05 | 70.33 | |
| LVRModel Category=Latent reasoning LVLMs2026.05 | 70.33 | |
| Gemini2.5-ProSize=-2026.06 | 69.8 | |
| Visual Para-Thinker++Size=3B2026.06 | 69.5 | |
| Majority voting@4Size=7B2026.06 | 68.9 | |
| GPT-4oModel Category=General-purpose LVLMs2026.05 | 68.7 | |
| Sequential (SFT)Size=7B2026.06 | 68.7 | |
| Qwen2.5-VLSize=7B2026.06 | 68.3 | |
| Emu3-ChatType=Und. Only, # LLM Params=8B2026.01 | 67 | |
| GPT5-miniSize=-2026.06 | 65.3 | |
| Qwen2.5-VLType=Und. Only, # LLM Params=3B2026.01 | 64.67 | |
| Claude-4-SonnetSize=-2026.06 | 63.9 | |
| GPT-4oSize=-2026.06 | 63.8 | |
| Visual Para-ThinkerSize=3B2026.06 | 63.6 | |
| Janus-ProType=Unified, # LLM Params=7B2026.01 | 63 | |
| GRPOSize=3B2026.06 | 62.1 | |
| Majority voting@4Size=3B2026.06 | 60.1 | |
| HarmonType=Unified, # LLM Params=1.5B2026.01 | 60 | |
| Sequential (SFT)Size=3B2026.06 | 60 | |
| Qwen2.5-VLSize=3B2026.06 | 59.2 | |
| JanusType=Unified, # LLM Params=1.3B2026.01 | 56.67 | |
| BAGELType=Unified, # LLM Params=3B2026.01 | 54.7 | |
| PAPOModel Category=RL-based reasoning LVLMs2026.05 | 54.33 | |
| LLaVA-OVType=Und. Only, # LLM Params=0.5B2026.01 | 53.67 | |
| Show-oType=Unified, # LLM Params=1.3B2026.01 | 50 | |
| ChameleonType=Unified, # LLM Params=7B2026.01 | 0 |