List your company
ResearchFormerly WizWand

Video Reasoning on Video-Holmes

46.7ScoreGPT-5
27.04432.14737.2542.353Aug 5, 2025Sep 3, 2025Oct 2, 2025Oct 31, 2025Nov 29, 2025Dec 28, 2025Jan 27, 2026
Updated 7mo ago

Evaluation results

MethodLinks
GPT-5Size=-, # Frames=-2026.0146.7
Qwen2.5-OmniRL Method=AVATAR2025.0845.1
Gemini-2.5-ProSize=-, # Frames=-2026.0145
Omni-R12025.0844.2
Qwen2.5-OmniRL Method=GRPO2025.0843.2
Video-KTRSize=7B, # Frames=642026.0142.7
Echolnk2025.0842.5
Ola-7BRL Method=AVATAR2025.0842.4
GPT-4oSize=-, # Frames=-2026.0142
Video-KTRSize=7B, # Frames=322026.0141.6
Our Final ModelTraining Stages=PT, RL, Decoding Mode=concise reason2025.1241.6
Gemini-1.5-ProSize=-, # Frames=-2026.0141.3
Ola-7BRL Method=GRPO2025.0841.3
Video-R12025.0841
Video-RTSSize=7B, # Frames=51.22026.0140.7
Video-KTRSize=7B, # Frames=162026.0140.7
Qwen3-VL-8BTraining Stages=PT, Decoding Mode=direct answer2025.1240.7
Qwen2.5-OmniRL Method=Baseline2025.0840.6
VideoRFT2025.0840.5
Ola-7BRL Method=Baseline2025.0840.1
HumanOmni2025.0839.6
AV-Reasoner2025.0839.6
Video-R1-7BTraining Stages=PT, SFT, RL, Decoding Mode=chain-of-thought2025.1239.4
TW-GRPO2025.0838.4
Qwen2.5-VL2025.0838
Qwen3-VL-8BTraining Stages=PT, Decoding Mode=concise reason2025.1237.4
Video-R1Size=7B, # Frames=322026.0136.5
Qwen2.5-VL-7BTraining Stages=PT, Decoding Mode=direct answer2025.1235.7
Qwen2.5-VL-SFTSize=7B, # Frames=322026.0133.9
Qwen2.5-VL-SFTSize=7B, # Frames=642026.0133.7
TW-GRPOSize=7B, # Frames=162026.0132.9
Qwen2.5-VL-7BTraining Stages=PT, Decoding Mode=concise reason2025.1232.1
Qwen2.5-VL-SFTSize=7B, # Frames=162026.0131.7
Qwen2.5-VLSize=7B, # Frames=-2026.0127.8