Multimodal Understanding on MME-RealWorld Lite
67.3Overall ScoreInternVL3-38B
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| InternVL3-38Bnumber of parameters=38B2026.02 | 67.3 | — | — | |
| InternVL3-78Bnumber of parameters=78B2026.02 | 65.4 | — | — | |
| TikArtParams=8B2026.02 | 56.97 | 60.48 | 53.47 | |
| GPT-52026.02 | 56.22 | 56.94 | 55.47 | |
| Monet-7BParams=7B2026.02 | 55.5 | 58.34 | 51.07 | |
| SkiLa2025.12 | 54.1 | 56.6 | 50.2 | |
| TikArt (w/o Observation)Params=8B2026.02 | 53.67 | 58.43 | 48.91 | |
| Gemini-2.5-pro2026.02 | 53.48 | 52.1 | 50.73 | |
| TikArt (w/o Zoom Action)Params=8B2026.02 | 52.56 | 57.91 | 47.2 | |
| Gemini-2.5-flash2026.02 | 52.07 | 49.7 | 47.32 | |
| GPT-4o2025.12 | 52 | 54.4 | 48.3 | |
| GPT-4oPrecision=BF16, Batch size=12026.06 | 52 | 54.4 | 48.3 | |
| Qwen3-VL-InstructParams=235B, MoE=true2026.02 | 51.82 | 56.3 | 60.78 | |
| DeepeyesParams=7B2026.02 | 50.79 | 55.3 | 46.29 | |
| TikArt (w/o Segment Action)Params=8B2026.02 | 50.69 | 53.51 | 47.87 | |
| SkiLa-V2025.12 | 50.2 | 51.7 | 47.7 | |
| EVA-VL 7BPrecision=BF16, Batch size=1, Model Parameter Scale=7B2026.06 | 49.8 | 53.3 | 44.4 | |
| LVR 7B*Model Size=7B, Tested by authors=true2025.12 | 49.1 | 52.7 | 43.5 | |
| TikArt (w/o GRPO)Params=8B2026.02 | 48.6 | 50.65 | 46.55 | |
| LLaVA-OneVision 7BModel Size=7B2025.12 | 48.5 | 52.8 | 42.7 | |
| Vision-R1 7B*Model Size=7B, Tested by authors=true2025.12 | 47.9 | 49.8 | 44.9 | |
| Direct SFT2025.12 | 47.5 | 49.6 | 44.3 | |
| Qwen2.5-VL 32BPrecision=BF16, Batch size=1, Model Parameter Scale=32B2026.06 | 46.2 | 50.6 | 39.3 | |
| Qwen2.5-VL-32Bnumber of parameters=32B2026.02 | 45.96 | — | — | |
| GPT-4o2026.02 | 45.73 | 49.81 | 53.89 | |
| TPRU-7Bnumber of parameters=7B2026.02 | 45.34 | — | — | |
| TPRU-32Bnumber of parameters=32B2026.02 | 44.92 | — | — | |
| Qwen2.5-VL-7Bnumber of parameters=7B2026.02 | 44.55 | — | — | |
| Qwen2.5-VL 7BModel Size=7B2025.12 | 44.3 | 48.6 | 37.7 | |
| Qwen2.5-VL 7BPrecision=BF16, Batch size=1, Model Parameter Scale=7B2026.06 | 44.1 | 48.8 | 37.7 | |
| GPT-5-nano2026.02 | 42.82 | 46.11 | 49.4 | |
| Qwen2.5-VL-3Bnumber of parameters=3B2026.02 | 41.94 | — | — | |
| Qwen3-VL-InstructParams=8B2026.02 | 41.64 | 49.02 | 34.27 | |
| Qwen3-VL-ThinkingParams=8B2026.02 | 39.43 | 44.05 | 34.8 | |
| TPRU-3Bnumber of parameters=3B2026.02 | 39.19 | — | — | |
| Qwen3-VL-InstructParams=32B2026.02 | 39.07 | 45.07 | 51.07 | |
| LLaVA-OneVision-1.5Params=8B2026.02 | 37.61 | 38.69 | 36.53 | |
| GPT-4o-mini2025.12 | 37.4 | 38.8 | 35.2 |