List your company

Spatio-temporal reasoning on V-STAR (test)

64.1What AccuracyOpen-o3 + MCoT
14.49227.37140.2553.129Mar 19, 2026Apr 4, 2026Apr 20, 2026May 6, 2026May 22, 2026Jun 7, 2026Jun 23, 2026

Evaluation results

MethodLinks
Open-o3 + MCoTGrounded reasoning component (r_ground)=true, Multi-Chain-of-Thought (MCoT)=true2026.0364.127.326.833.638.136.650.5
Open-o3 + MCoTGrounded reasoning component (r_ground)=false, Multi-Chain-of-Thought (MCoT)=true2026.0363.426.325.630.531.435.249.1
Motion-oGrounded reasoning component (r_ground)=true2026.0362.726.626.228.215.135.549.4
Motion-oGrounded reasoning component (r_ground)=false2026.0362.226.225.826.612.33548.8
SER-7BParameters=7B2026.0661.627.927.330.75.135.749.6
Open-o3 Video2026.036124.52425.4633.746.6
Open-o3-VideoReference=ICML262026.066124.52425.4633.746.6
GPT-4o2026.0360.816.712.86.5326.838.2
GPT-4o2026.0660.816.712.86.5326.838.2
Gemini-2-Flash2026.035324.523.84.62.226.935.6
Gemini-2-Flash2026.065324.523.84.62.226.935.6
LLaVA-Video2026.0349.510.512.21.91.320.827.3
LLaVA-Video2026.0649.510.512.21.91.320.827.3
InternVL-2.5-8B2026.0344.28.77.80.70.117.624.9
InternVL-2.5-8BParameters=8B2026.0644.28.77.80.70.117.624.9
Video-LLaMA32026.0341.92323.10.90.221.727
Video-LLaMA32026.0641.92323.10.90.221.727
VideoChat22026.0336.213.712.52.511720.3
VideoChat22026.0636.213.712.52.511720.3
Qwen2.5-VL-7B (base)Evaluation framework=vLLM, Number of sampled frames=162026.0333.515.413.8172.519.322.4
Qwen2.5-VL-7BParameters=7B, Configuration=Base model2026.0633.515.413.8172.519.322.4
Oryx-1.5-7B2026.0320.513.514.810.13.515.113.8
Oryx-1.5-7BParameters=1.5-7B2026.0620.513.514.810.13.515.113.8
TRACE2026.0317.619.117.1001213.3
TRACE2026.0617.619.117.1001213.3
Sa2VA-8B2026.0316.40.1032.337.517.120.3
Sa2VA-8BParameters=8B2026.0616.40.1032.337.517.120.3