Autonomous Driving Reasoning on DriveLM
81AccuracyDriveMM
Evaluation Results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| DriveMMTokens per image=729, Pruning ratio=0%, Base Model=DriveMM2026.06 | 81 | 65.44 | — | 33.9 | 59.1 | — | — | — | — | 61 | 74 | 19 | |
| Prune2DriveTokens per image=180, Pruning ratio=75%, Base Model=DriveMM2026.06 | 80 | 64.92 | — | 34 | 58.3 | — | — | — | — | 60 | 75 | 20 | |
| SparseVLMTokens per image=180, Pruning ratio=75%, Base Model=DriveMM2026.06 | 79 | 63.74 | — | 33.4 | 57.4 | — | — | — | — | 57 | 73 | 19 | |
| DivPruneTokens per image=180, Pruning ratio=75%, Base Model=DriveMM2026.06 | 79 | 64.67 | — | 33.7 | 57.9 | — | — | — | — | 59 | 74 | 21 | |
| MVPrunerTokens per image=180, Pruning ratio=75%, Base Model=DriveMM2026.06 | 79 | 65.52 | — | 34.3 | 58.7 | — | — | — | — | 61 | 75 | 22 | |
| MVPrunerTokens per image=72, Pruning ratio=90%, Base Model=DriveMM2026.06 | 79 | 65.41 | — | 34 | 58.2 | — | — | — | — | 58 | 74 | 17 | |
| DARTTokens per image=180, Pruning ratio=75%, Base Model=DriveMM2026.06 | 78 | 64.02 | — | 33.6 | 57.2 | — | — | — | — | 56 | 73 | 18 | |
| PACTTokens per image=180, Pruning ratio=75%, Base Model=DriveMM2026.06 | 78 | 64.24 | — | 33.3 | 57.3 | — | — | — | — | 56 | 73 | 19 | |
| DivPruneTokens per image=72, Pruning ratio=90%, Base Model=DriveMM2026.06 | 78 | 64.16 | — | 32.7 | 57 | — | — | — | — | 55 | 73 | 17 | |
| Prune2DriveTokens per image=72, Pruning ratio=90%, Base Model=DriveMM2026.06 | 78 | 64.52 | — | 33.4 | 57.4 | — | — | — | — | 56 | 74 | 16 | |
| VGGDriveEvaluation Protocol=Fine-tuned2026.02 | 77.5 | 64.76 | 56.69 | 49.77 | 61.26 | — | — | — | — | — | — | — | |
| FastVTokens per image=180, Pruning ratio=75%, Base Model=DriveMM2026.06 | 77 | 63.95 | — | 32.8 | 56.6 | — | — | — | — | 54 | 72 | 15 | |
| PACTTokens per image=72, Pruning ratio=90%, Base Model=DriveMM2026.06 | 76 | 64.76 | — | 32.9 | 56.8 | — | — | — | — | 54 | 73 | 15 | |
| SparseVLMTokens per image=72, Pruning ratio=90%, Base Model=DriveMM2026.06 | 75 | 63.38 | — | 32.9 | 55.9 | — | — | — | — | 52 | 72 | 15 | |
| I-VL4DriveEvaluation Protocol=Fine-tuned2026.02 | 73.39 | 65.25 | 48.56 | 47.65 | 60.02 | — | — | — | — | — | — | — | |
| FSDriveEvaluation Protocol=Fine-tuned2026.02 | 71.77 | 63.42 | 52.77 | 39.19 | 57.12 | — | — | — | — | — | — | — | |
| OminiDriveEvaluation Protocol=Fine-tuned2026.02 | 70 | 65 | — | 37 | 56 | — | — | — | — | — | — | — | |
| DARTTokens per image=72, Pruning ratio=90%, Base Model=DriveMM2026.06 | 69 | 64.2 | — | 32 | 54.7 | — | — | — | — | 51 | 71 | 12 | |
| FastVTokens per image=72, Pruning ratio=90%, Base Model=DriveMM2026.06 | 68 | 63.52 | — | 32.3 | 54.1 | — | — | — | — | 49 | 71 | 8 | |
| BaselineEvaluation Protocol=Fine-tuned2026.02 | 64.35 | 65.3 | 44.9 | 34.54 | 54.59 | — | — | — | — | — | — | — | |
| TM-LMMEvaluation Protocol=Fine-tuned2026.02 | 59.6 | 58.44 | 46.38 | 35.73 | 51.5 | — | — | — | — | — | — | — | |
| RoboTronEvaluation Protocol=Zero-shot2026.02 | 52.94 | 51.11 | 7.12 | 35.05 | 39.47 | — | — | — | — | — | — | — | |
| DriveLMEvaluation Protocol=Fine-tuned2026.02 | 52.3 | 55.35 | 45.13 | 35.73 | 47.2 | — | — | — | — | — | — | — | |
| GPT-4oEvaluation Protocol=Zero-shot2026.02 | 38.55 | 67.27 | 8.97 | 24 | 41.21 | — | — | — | — | — | — | — | |
| LLAVA-OVEvaluation Protocol=Zero-shot2026.02 | 25.03 | 65.7 | 14.44 | 40.93 | 42.36 | — | — | — | — | — | — | — | |
| AD-MLLM + CMSafeguard Method=CM2026.05 | — | — | — | — | — | — | 58.6 | 32.87 | — | — | — | — | |
| AD-MLLM + RFSafeguard Method=RF2026.05 | — | — | — | — | — | — | 58.78 | 32.67 | — | — | — | — | |
| AD-MLLM + SASafeguard Method=SA2026.05 | — | — | — | — | — | — | 60.08 | 35.95 | — | — | — | — | |
| DriveLMM-o1Params=8B2025.11 | — | — | — | — | — | 37.05 | — | — | — | — | — | — | |
| DriveMMPrefilling Latency (ms)=1968, Decoding Latency (ms)=65.3, FLOPs=100%, GPU Peak Memory (GB)=16.47, Throughput (samples/s)=8.12, Pruning Setting=None2026.06 | — | — | — | — | — | — | — | — | 59.1 | — | — | — | |
| FastVPrefilling Speedup (ms)=340.5(↑5.78×), Decoding Speedup (ms)=63.4(↑1.03×), FLOPs=13.3%, GPU Peak Memory (GB)=15.99, Throughput (samples/s)=5.56(↑1.46×), Pruning Setting=90%2026.06 | — | — | — | — | — | — | — | — | 55.4 | — | — | — | |
| Gemini1.5-ProParams=–2025.11 | — | — | — | — | — | 39.92 | — | — | — | — | — | — | |
| GPT-4oParams=–2025.11 | — | — | — | — | — | 41.21 | — | — | — | — | — | — | |
| GUARDADSafeguard Method=GA2026.05 | — | — | — | — | — | — | 61.36 | 37.51 | — | — | — | — | |
| InternVL2-26BParams=38B2025.11 | — | — | — | — | — | 24.57 | — | — | — | — | — | — | |
| InternVL2-L-1.5Params=8B2025.11 | — | — | — | — | — | 29.74 | — | — | — | — | — | — | |
| MiMo-EmbodiedParams=7B2025.11 | — | — | — | — | — | 57.85 | — | — | — | — | — | — | |
| MiMo-VLParams=7B2025.11 | — | — | — | — | — | 29.76 | — | — | — | — | — | — | |
| MVPrunerPrefilling Speedup (ms)=395.6(↑4.97×), Decoding Speedup (ms)=51.8(↑1.26×), FLOPs=12.7%, GPU Peak Memory (GB)=15.93, Throughput (samples/s)=5.58(↑1.45×), Pruning Setting=90%2026.06 | — | — | — | — | — | — | — | — | 58.2 | — | — | — | |
| Prune2DrivePrefilling Speedup (ms)=307.5(↑6.40×), Decoding Speedup (ms)=59.9(↑1.09×), FLOPs=13.4%, GPU Peak Memory (GB)=16.00, Throughput (samples/s)=5.49(↑1.48×), Pruning Setting=90%2026.06 | — | — | — | — | — | — | — | — | 57.4 | — | — | — | |
| Qwen-VL-MaxParams=–2025.11 | — | — | — | — | — | 26.99 | — | — | — | — | — | — | |
| Qwen2-VLParams=7B2025.11 | — | — | — | — | — | 25.39 | — | — | — | — | — | — | |
| Qwen2-VLParams=72B2025.11 | — | — | — | — | — | 29.7 | — | — | — | — | — | — | |
| RoboTron-DriveParams=8B2025.11 | — | — | — | — | — | 61.3 | — | — | — | — | — | — | |
| SparseVLMPrefilling Speedup (ms)=484.7(↑4.06×), Decoding Speedup (ms)=64.0(↑1.02×), FLOPs=14.4%, GPU Peak Memory (GB)=16.00, Throughput (samples/s)=5.88(↑1.38×), Pruning Setting=90%2026.06 | — | — | — | — | — | — | — | — | 55.9 | — | — | — | |
| Specialist ModelParams=–2025.11 | — | — | — | — | — | 57 | — | — | — | — | — | — | |
| Vanilla AD-MLLMSafeguard Method=Vanilla2026.05 | — | — | — | — | — | — | 58.39 | 32.47 | — | — | — | — |