Assistive Vision-Language Tasks on benchmark local images disk-replay protocol (n=120) v6
518TTFTMiniCPM-o 4.5
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| MiniCPM-o 4.5Deployment Environment=Local on-device (local inference)2026.07 | 518 | 682 | 1.283 | 90.8 | 31.7 | |
| MiniCPM-V 4.5Deployment Environment=Local on-device (local inference)2026.07 | 526 | 692 | 1.325 | 89.2 | 37.5 | |
| MiniCPM-o 4.5 ThinkDeployment Environment=Local on-device (local inference)2026.07 | 544 | 710 | 1.017 | 85 | 21.7 | |
| Qwen-VL-MaxDeployment Environment=Cloud-based (remote inference)2026.07 | 616 | 984 | 1.408 | 93.3 | 30 | |
| NaiveOnDeviceDeployment Environment=Local on-device (local inference)2026.07 | 1,380 | 1,381 | 1.333 | 90 | 32.5 | |
| Gemini 2.5 FlashDeployment Environment=Cloud-based (remote inference)2026.07 | 5,428 | 5,441 | 1.425 | 95.8 | 31.7 |