List your company
ResearchFormerly WizWand

Spatiotemporal-sensitive video understanding on DyBench

82.2AccuracyGemini-3.1-Pro
47.04856.17465.374.426May 21, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Gemini-3.1-ProModel Type=Proprietary2026.0582.271.7
Qwen3-VL-8B + CRPOBackbone=Qwen3-VL-8B, Post-training Method=CRPO2026.0572.558.1
Qwen3-VL-8B + T-GRPOBackbone=Qwen3-VL-8B, Post-training Method=T-GRPO2026.0570.954.5
Qwen3-VL-4B + CRPOBackbone=Qwen3-VL-4B, Post-training Method=CRPO2026.0570.354.8
Qwen3-VL-8B + ArrowRLBackbone=Qwen3-VL-8B, Post-training Method=ArrowRL2026.0569.652.2
InternVL3-8BModel Type=Open-Source2026.0569.550.2
Qwen3-VL-8B + GRPOBackbone=Qwen3-VL-8B, Post-training Method=GRPO2026.0569.452.4
Qwen3-VL-4B + T-GRPOBackbone=Qwen3-VL-4B, Post-training Method=T-GRPO2026.0568.649.8
Qwen3-VL-8BBackbone=Qwen3-VL-8B, Post-training Method=base2026.0568.250.4
ArrowRL*Model Type=Open-Source2026.0567.550.1
Qwen3-VL-4B + ArrowRLBackbone=Qwen3-VL-4B, Post-training Method=ArrowRL2026.0566.747.5
Qwen3-VL-4B + GRPOBackbone=Qwen3-VL-4B, Post-training Method=GRPO2026.0566.648.2
Qwen3-VL-4BBackbone=Qwen3-VL-4B, Post-training Method=base2026.0565.145.4
Qwen2.5-VL-7BModel Type=Open-Source2026.0564.544.1
GPT-5.1Model Type=Proprietary2026.0563.744.9
LLaVA-OV-7BModel Type=Open-Source2026.0548.419.6