Real-world visual perception on MME-RealWorld Lite
56.38AccuracyKimi-K2.5
Evaluation Results
| Method | Links | |
|---|---|---|
| Kimi-K2.5Param Size=1T, Inference Mode=Single Forward Pass2026.05 | 56.38 | |
| Vision-OPD (Ours)Param Size=4B, Inference Mode=Single Forward Pass2026.05 | 55.45 | |
| ThymeParam Size=7B, Inference Mode=Agentic2026.05 | 55.18 | |
| ZwZParam Size=8B, Inference Mode=Single Forward Pass2026.05 | 54.35 | |
| DeepEyesParam Size=7B, Inference Mode=Agentic2026.05 | 53.2 | |
| DeepEyesV2Param Size=7B, Inference Mode=Agentic2026.05 | 51.59 | |
| Vision-OPD (Ours)Param Size=9B, Inference Mode=Single Forward Pass2026.05 | 49.56 | |
| Gemini-3.1-ProParam Size=-, Inference Mode=Closed-Source2026.05 | 49.5 | |
| GPT-5.4Param Size=-, Inference Mode=Closed-Source2026.05 | 48.1 | |
| GPT-5.2Param Size=-, Inference Mode=Closed-Source2026.05 | 47.06 | |
| Qwen3-VL-InstructParam Size=8B, Inference Mode=Single Forward Pass2026.05 | 46.53 | |
| Qwen3.5Param Size=9B, Inference Mode=Single Forward Pass2026.05 | 46.12 | |
| Qwen3.5Param Size=397B, Inference Mode=Single Forward Pass2026.05 | 45.39 | |
| SenseNova-MARSParam Size=8B, Inference Mode=Agentic2026.05 | 43.67 | |
| GLM-4.6VParam Size=106B, Inference Mode=Single Forward Pass2026.05 | 43.1 | |
| Qwen3-VL-InstructParam Size=235B, Inference Mode=Single Forward Pass2026.05 | 42.83 | |
| MiniCPM-V-4.5Param Size=9B, Inference Mode=Single Forward Pass2026.05 | 42.31 | |
| MiMo-VL-RLParam Size=7B, Inference Mode=Single Forward Pass2026.05 | 39.5 | |
| Qwen3.5Param Size=4B, Inference Mode=Single Forward Pass2026.05 | 35.8 |