List your company

Autonomous Driving Reasoning on DriveLMM-o1

82.31Risk AssessmentOmniDrive-R1
45.005254.690164.37574.0599Dec 16, 2025Jan 18, 2026Feb 21, 2026Mar 26, 2026Apr 29, 2026Jun 1, 2026Jul 5, 2026

Evaluation results

MethodLinks
OmniDrive-R1reasoning_strategy=Multi-modal Interleaved Chain-of-Thought2025.1282.3185.4283.7582.5878.2680.3573.62—————
OmniDrive-R12026.0782.3185.4283.7582.5878.2680.3573.62—————
CritiqueDriveVLM-TeacherStage=22026.0781.6386.2182.2785.4679.1780.4876.54—————
Agentthink2025.1280.5184.9882.1184.9979.5679.6871.35—————
AgentThink2026.0780.5184.9882.1184.9979.5679.6871.35—————
Qwen3-VLParameters=8B2026.0779.584.3280.4184.4375.3277.7655.54—————
DriveLMM-o1Tokens retained per image=256, Retention percentage=100%2026.0673.8182.2675.6479.5171.3874.37——————
DriveLMM-o12025.1273.0181.5675.3979.4274.4975.2462.36—————
DriveLMM-o12026.0773.0181.5675.3979.4274.4975.2462.36—————
MVPrunerTokens retained per image=64, Retention percentage=25%2026.0672.9181.9674.8578.5571.5873.79——————
MVPrunerTokens retained per image=25, Retention percentage=10%2026.0672.0581.8173.877.6170.2572.91——————
DivPruneTokens retained per image=64, Retention percentage=25%, Venue=CVPR252026.0671.6678.3973.0477.4569.9271.67——————
DARTTokens retained per image=64, Retention percentage=25%, Venue=EMNLP252026.0671.578.2172.877.3369.6371.46——————
FastVTokens retained per image=64, Retention percentage=25%, Venue=ECCV242026.0671.3580.5872.8576.7469.2572.41——————
GPT-4o2025.1271.3280.7272.9676.6571.4372.5257.84—————
GPT-4o2026.0771.3280.7272.9676.6571.4372.5257.84—————
Prune2DriveTokens retained per image=64, Retention percentage=25%, Venue=CVPR262026.0671.3180.7473.0777.4869.372.69——————
PACTTokens retained per image=64, Retention percentage=25%, Venue=CVPR252026.0671.278.1171.4477.2168.9570.95——————
Prune2DriveTokens retained per image=25, Retention percentage=10%, Venue=CVPR262026.0670.7677.6672.2977.0168.9470.31——————
SparseVLMTokens retained per image=64, Retention percentage=25%, Venue=ICML252026.0670.0879.5971.8775.6667.4371.23——————
DivPruneTokens retained per image=25, Retention percentage=10%, Venue=CVPR252026.0669.3276.4770.7875.5368.2269.65——————
InternVL2.5-8BParameters=8B2025.1269.0278.4371.5275.870.5471.6254.87—————
InternVL2.5Parameters=8B2026.0769.0278.4371.5275.870.5471.6254.87—————
DARTTokens retained per image=25, Retention percentage=10%, Venue=EMNLP252026.0668.675.8169.9774.9467.669.01——————
PACTTokens retained per image=25, Retention percentage=10%, Venue=CVPR252026.0668.276.2369.1174.1466.168.38——————
SparseVLMTokens retained per image=25, Retention percentage=10%, Venue=ICML252026.0668.1877.8169.8773.9866.269.4——————
FastVTokens retained per image=25, Retention percentage=10%, Venue=ECCV242026.0667.7478.0769.3773.2466.0269.56——————
Qwen2.5VL-72BParameters=72B2025.1264.472.8160.2965.1362.8165.7361.27—————
Qwen2.5-VLParameters=72B2026.0764.472.8160.2965.1362.8165.7361.27—————
LlamaV-O12025.1260.273.5262.6764.6663.4163.1350.02—————
LlamaV-O12026.0760.273.5262.6764.6663.4163.1350.02—————
LLaVA-CoT2025.1257.6269.0160.8462.7260.6761.4149.27—————
LLaVA-CoT2026.0757.6269.0160.8462.7260.6761.4149.27—————
DriveLMM-o1LMM=Qwen2.5-VL2026.0353.0868.5249.91——56.6351.1447.4350.9651.7——
RailVQA-CoMLMM=DriveLMM-o1(Qwen2.5-VL)2026.0352.8868.0756.48——58.8954.5748.6652.8153.77——
DriveLMM-o1LMM=InternVL-2.52026.0352.5567.2254.67——57.4854.1149.8451.251.69——
Mulberry-7BParameters=7B2025.1251.8963.6656.6857.2757.4557.6552.86—————
MulberryParameters=7B2026.0751.8963.6656.6857.2757.4557.6552.86—————
Ovis1.5-Gemma2-9BParameters=9B2025.1251.3466.3654.7455.7255.7455.6248.85—————
Ovis1.5-Gemma2Parameters=9B2026.0751.3466.3654.7455.7255.7455.6248.85—————
Qwen2.5VL-7BParameters=7B2025.1246.4460.4551.0250.1552.1951.7737.81—————
Qwen2.5-VLParameters=7B2026.0746.4460.4551.0250.1552.1951.7737.81—————
Cosmos-R1Category=Embodied Models2026.04——————————54.66—
DriveLMM-o1Prefilling Latency (ms)=591, Decoding Latency (ms)=51.9, FLOPs=100%, GPU Peak Memory (GB)=17.31, Throughput (samples/s)=11.84, Pruning Setting=None2026.06———————————74.37
DriveMMCategory=Embodied Models2026.04——————————65.91—
FastVPrefilling Speedup (ms)=128.5(↑4.60×), Decoding Speedup (ms)=45.9(↑1.13×), FLOPs=20.3%, GPU Peak Memory (GB)=16.61, Throughput (samples/s)=10.57(↑1.12×), Pruning Setting=90%2026.06———————————69.56
GPT-4oCategory=Proprietary Models2026.04——————————57.84—
Mimo-EmbodiedCategory=Embodied Models2026.04——————————40.31—
MVPrunerPrefilling Speedup (ms)=155.5(↑3.80×), Decoding Speedup (ms)=42.2(↑1.23×), FLOPs=20.0%, GPU Peak Memory (GB)=16.08, Throughput (samples/s)=8.60(↑1.38×), Pruning Setting=90%2026.06———————————72.91
PR1-Counting-2BCategory=Spatial Models2026.04——————————50.94—
PR1-Detection-3BCategory=Spatial Models2026.04——————————44.81—
PR1-Grounding-2BCategory=Spatial Models2026.04——————————53.83—
PR1-OCR-2BCategory=Spatial Models2026.04——————————47.68—
Prune2DrivePrefilling Speedup (ms)=127.1(↑4.65×), Decoding Speedup (ms)=47.2(↑1.10×), FLOPs=20.3%, GPU Peak Memory (GB)=16.58, Throughput (samples/s)=8.58(↑1.38×), Pruning Setting=90%2026.06———————————70.31
Qwen2.5-VL-32BCategory=Open-Source Models2026.04——————————55.04—
Qwen2.5-VL-7BCategory=Open-Source Models2026.04——————————37.81—
Qwen3-VL-32BCategory=Open-Source Models2026.04——————————55.29—
Qwen3-VL-A3B-30BCategory=Open-Source Models2026.04——————————57.17—
SparseVLMPrefilling Speedup (ms)=164.6(↑3.59×), Decoding Speedup (ms)=47.6(↑1.09×), FLOPs=20.7%, GPU Peak Memory (GB)=16.65, Throughput (samples/s)=9.55(↑1.24×), Pruning Setting=90%2026.06———————————69.4
UniVG-R1-7BCategory=Spatial Models2026.04——————————51.42—
XEmbodiedversion=Best2026.04——————————77.01—