List your company
ResearchFormerly WizWand

Audio-visual understanding on Daily-Omni

82.8AccuracyOmniVideo-R1
26.057640.788855.5270.2512Dec 11, 2025Jan 11, 2026Feb 12, 2026Mar 15, 2026Apr 16, 2026May 17, 2026Jun 18, 2026
Updated 4mo ago

Evaluation results

MethodLinks
OmniVideo-R1Model access type=Open-source2026.0282.8
Qwen3.5-OmniOpen-Source=✗, Size=Flash2026.0481.8
Gemini-2.5-ProModel access type=Closed-source2026.0281.4
Gemini-3-ProModel access type=Closed-source2026.0281.1
video-SALMONN 2+-72BModel access type=Open-source2026.0279.4
Qwen3-Omni-30B-A3B-ThinkingModel access type=Open-source2026.0275.8
Nemotron 3 Nano OmniOpen-Source=✓, Size=30B-A3B, Mode=Reasoning off2026.0474.5
CogniRoute2026.0674.3
Nemotron 3 Nano OmniOpen-Source=✓, Size=30B-A3B, Mode=Reasoning on2026.0474.1
Qwen3-OmniOpen-Source=✓, Size=30B-A3B, Mode=Thinking2026.0473.6
Qwen3 Omni 30BParameters=30B2026.0673.6
Qwen3-OmniOpen-Source=✓, Size=30B-A3B, Mode=Instruct2026.0471.9
video-SALMONN 2+-7BModel access type=Open-source2026.0271.8
DPO w/ SPRecipe=DPO with SFT-policy negatives2026.0569
DPO w/ OP + SPRecipe=DPO with original-sync and SFT-policy negatives2026.0568.5
Qwen3-Omni-30BRecipe=Vanilla Baseline2026.0568.2
DPO w/ SP + FV-DRecipe=DPO with SFT-policy negatives and video preference data2026.0568
DPO w/ CTP + FV-DRecipe=DPO with counterfactual and video preference data2026.0568
Final 10K DPO Recipe (Ours)Recipe=Combined CTP, FV-D, and FV-A-L2026.0567.9
Gemini-2.0-FlashModel access type=Closed-source2026.0267.8
DPO w/ CTP + FV-D + LV-MCQARecipe=DPO with counterfactual, video preference, and MCQA data2026.0567.8
DPO w/ OP + FV-D + LV-MCQARecipe=DPO with original-sync, video preference, and MCQA data2026.0567.6
DPO w/ CTP + FV-D + FV-ARecipe=DPO with counterfactual, video preference, and audio preference data2026.0567.3
SFT w/ CTP + FV-D + FV-ALRecipe=SFT with counterfactual, general video preference, and audio-visual data2026.0566.9
v-SALMONN-o1Reasoning=true, Model Size=7B, Zero-shot=true2026.0464
Qwen3-Omni-30B-A3B-InstructModel access type=Open-source2026.0263.6
EchoingPixelsModel Size=3B, Token Budget=20%, Est. Tokens Per Min=2,0282025.1260.65
Qwen2.5-Omni-7BModel Size=7B, Est. Tokens Per Min=10,1402025.1260.48
Qwen2.5-Omni-3BModel Size=3B, Est. Tokens Per Min=10,1402025.1259.65
HumanOmniV2-7BModel access type=Open-source2026.0258.5
EchoingPixelsModel Size=3B, Token Budget=10%, Est. Tokens Per Min=1,0142025.1257.56
Qwen3-OmniReasoning=true, Model Size=30B, Zero-shot=true2026.0457.5
EchoingPixelsModel Size=7B, Token Budget=10%, Est. Tokens Per Min=1,0142025.1256.98
AVRTReasoning=true, Model Size=7B, Zero-shot=true2026.0454.4
AV-ReasonerReasoning=true, Model Size=7B, Zero-shot=true2026.0453.8
FastVModel Size=7B, Token Budget=20%, Est. Tokens Per Min=2,0282025.1253.55
MiniCPM-o-7BModel access type=Open-source2026.0253.1
EchoingPixelsModel Size=3B, Token Budget=5%, Est. Tokens Per Min=5072025.1252.88
AF3 (think)Reasoning=true, Model Size=Teacher, Zero-shot=true2026.0452.5
Ola-7BReasoning=false, Model Size=7B, Zero-shot=true2026.0452.3
Qwen2.5 OmniReasoning=false, Model Size=7B, Zero-shot=true2026.0451.5
OlaModel Size=7B, Est. Tokens Per Min=∼16,0002025.1250.71
Ola-7BModel access type=Open-source2026.0249.9
FastVModel Size=3B, Token Budget=20%, Est. Tokens Per Min=2,0282025.1249.87
IntraModalModel Size=7B, Token Budget=25%, Est. Tokens Per Min=2,5352025.1249.79
AVRTReasoning=true, Model Size=3B, Zero-shot=true2026.0449.2
V-RTSReasoning=true, Model Size=7B, Zero-shot=true2026.0447.8
HumanOmniReasoning=true, Model Size=7B, Zero-shot=true2026.0447.6
Qwen2.5-Omni-7BModel access type=Open-source2026.0247.5
Kimi-VL-ThinkingReasoning=true, Model Size=Teacher, Zero-shot=true2026.0447
AVATARReasoning=true, Model Size=7B, Zero-shot=true2026.0447
Omni-R1Reasoning=true, Model Size=7B, Zero-shot=true2026.0446.8
IntraModalModel Size=3B, Token Budget=25%, Est. Tokens Per Min=2,5352025.1246.2
EchoInkReasoning=true, Model Size=7B, Zero-shot=true2026.0446.2
AVATARReasoning=true, Model Size=3B, Zero-shot=true2026.0444.7
Qwen2.5 OmniReasoning=false, Model Size=3B, Zero-shot=true2026.0443.1
VideoLLaMA2-7BModel access type=Open-source2026.0235.2
VideoLLaMA2Model Size=7B, Est. Tokens Per Min=∼4,1522025.1235.17
Unified-IO-2-XLModel Size=3B, Est. Tokens Per Min=8962025.1228.32
Unified-IO-2-XXLModel Size=7B, Est. Tokens Per Min=8962025.1228.24