List your company
ResearchFormerly WizWand

Multi-turn Dialogue on MT-Bench (MT-Bench Score)

76.19MT-Bench ScoreQwen3-Omni-30B-A3B-Thinking
3.161222.120641.0860.0394Oct 28, 2024Jan 30, 2025May 4, 2025Aug 6, 2025Nov 8, 2025Feb 10, 2026May 15, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Qwen3-Omni-30B-A3B-ThinkingModel Category=Large Sized Large Audio Language Models, Parameter Count=>20B parameters2025.0976.19
Gemini2.5-FlashModel Category=Proprietary Audio Language Models2025.0974.5
Whisper-Large-v3 + GPT-oSS-20BModel Category=Cascaded Systems2025.0971.5
Voxtral-Small-24BModel Category=Large Sized Large Audio Language Models, Parameter Count=>20B parameters2025.0970.81
GPT-4o-transcribe + GPT-4.1-miniModel Category=Cascaded Systems2025.0967.62
Voxtral-Mini-3BModel Category=Small-sized Audio Language Models, Parameter Count=<5B parameters2025.0965.88
GPT-4o-mini-audioModel Category=Proprietary Audio Language Models2025.0965
Qwen2.5-Omni-7BModel Category=Medium Sized Large Audio Language Models, Parameter Count=5B-20B parameters2025.0964.56
Phi-4-Multi-modalModel Category=Medium Sized Large Audio Language Models, Parameter Count=5B-20B parameters2025.0964.12
Qwen2.5-Omni-3BModel Category=Small-sized Audio Language Models, Parameter Count=<5B parameters2025.0959.81
Kimi-AudioModel Category=Medium Sized Large Audio Language Models, Parameter Count=5B-20B parameters2025.0954.62
ξ-DPOBackbone=Gemma2-Instruct (9B)2026.059
I-DPO + MaPPOModel=Qwen2.5-32B-Instruct2025.078.99
I-DPOModel=Qwen2.5-32B-Instruct2025.078.97
DPO + MaPPOModel=Qwen2.5-32B-Instruct2025.078.9
ITModel=Qwen2.5-32B-Instruct2025.078.89
DPO + TPO2026.058.81
I-DPO + MaPPOModel=Qwen2.5-14B-Instruct2025.078.8
DPO + Topo-TPO2026.058.8
DPO + MaPPOModel=Qwen2.5-7B-Instruct2025.078.79
DPO + MaPPOModel=Qwen2.5-14B-Instruct2025.078.68
Qwen+UFTPhase=Instruction-tuning and alignment data comparison2024.108.67
DPO2026.058.65
Qwen+SFT+DPOPhase=Instruction-tuning and alignment data comparison2024.108.64
I-DPOModel=Qwen2.5-14B-Instruct2025.078.64
ITModel=Qwen2.5-14B-Instruct2025.078.62
ITModel=Qwen2.5-7B-Instruct2025.078.61
Qwen+SFT+KTOPhase=Instruction-tuning and alignment data comparison2024.108.58
Qwen+SFT+UNAPhase=Instruction-tuning and alignment data comparison2024.108.57
DPOModel=Qwen2.5-7B-Instruct2025.078.56
I-DPOModel=Qwen2.5-7B-Instruct2025.078.55
I-DPO + MaPPOModel=Qwen2.5-7B-Instruct2025.078.54
SFTBackbone=Gemma2-Instruct (9B)2026.058.5
DPOBackbone=Gemma2-Instruct (9B)2026.058.5
DPOModel=Qwen2.5-32B-Instruct2025.078.47
DPOModel=Qwen2.5-14B-Instruct2025.078.46
IPOBackbone=Llama3-Instruct (8B)2026.058.3
ORPOBackbone=Llama3-Instruct v0.2 (8B)2026.058.3
KTOBackbone=Llama3-Instruct (8B)2026.058.2
ξ-DPOBackbone=Llama3-Instruct (8B)2026.058.2
DPOBackbone=Llama3-Instruct v0.2 (8B)2026.058.2
IPOBackbone=Llama3-Instruct v0.2 (8B)2026.058.2
CPOBackbone=Llama3-Instruct v0.2 (8B)2026.058.2
KTOBackbone=Llama3-Instruct v0.2 (8B)2026.058.2
R-DPOBackbone=Llama3-Instruct v0.2 (8B)2026.058.2
DPO + MaPPOModel=Llama-3-8B-Instruct2025.078.18
DPO + MaPPOModel=Qwen2.5-3B-Instruct2025.078.13
I-DPOModel=Qwen2.5-3B-Instruct2025.078.1
SFTBackbone=Llama3-Instruct (8B)2026.058.1
SFTBackbone=Llama3-Instruct v0.2 (8B)2026.058.1
DPOModel=Llama-3-8B-Instruct2025.078.07
I-DPO + MaPPOModel=Llama-3-8B-Instruct2025.078.04
DPOModel=Qwen2.5-3B-Instruct2025.078.02
I-DPOModel=Llama-3-8B-Instruct2025.078.01
I-DPO + MaPPOModel=Qwen2.5-3B-Instruct2025.078.01
DPOBackbone=Llama3-Instruct (8B)2026.058
CPOBackbone=Llama3-Instruct (8B)2026.058
ORPOBackbone=Llama3-Instruct (8B)2026.058
R-DPOBackbone=Llama3-Instruct (8B)2026.058
SimPOBackbone=Llama3-Instruct (8B)2026.058
SimPOBackbone=Llama3-Instruct v0.2 (8B)2026.058
ξ-DPOBackbone=Llama3-Instruct v0.2 (8B)2026.058
QwenPhase=Instruction-tuning data comparison2024.107.97
Qwen+UFTPhase=Instruction-tuning data comparison2024.107.95
ITModel=Qwen2.5-3B-Instruct2025.077.92
Qwen+SFTPhase=Instruction-tuning data comparison2024.107.85
IPOBackbone=Mistral-Instruct (7B)2026.057.8
KTOBackbone=Mistral-Instruct (7B)2026.057.7
ORPOBackbone=Mistral-Instruct (7B)2026.057.7
ξ-DPOBackbone=Mistral-Instruct (7B)2026.057.7
I-DPO + MaPPOModel=Qwen2.5-1.5B-Instruct2025.077.63
DPOBackbone=Mistral-Instruct (7B)2026.057.6
SimPOBackbone=Mistral-Instruct (7B)2026.057.6
I-DPO + MaPPOModel=Mistral-7B-Instruct2025.077.59
DPO + MaPPOModel=Qwen2.5-1.5B-Instruct2025.077.57
ITModel=Llama-3-8B-Instruct2025.077.52
DPO + MaPPOModel=Mistral-7B-Instruct2025.077.51
SFTBackbone=Mistral-Instruct (7B)2026.057.5
CPOBackbone=Mistral-Instruct (7B)2026.057.5
R-DPOBackbone=Mistral-Instruct (7B)2026.057.5
I-DPOModel=Qwen2.5-1.5B-Instruct2025.077.39
DPOModel=Qwen2.5-1.5B-Instruct2025.077.29
ITModel=Qwen2.5-1.5B-Instruct2025.077.06
I-DPOModel=Mistral-7B-Instruct2025.076.92
DPOModel=Mistral-7B-Instruct2025.076.86
LoRA+LLM=Llama 3.1-8B2026.056.35
LoRA-Over-MPO_RLLM=Llama 3.1-8B, rank=1282026.056.31
LoRA-Over-MPO_RLLM=Llama 3.1-8B2026.056.26
DoRALLM=Llama 3.1-8B2026.056.24
LoRA-Over-MPO_RLLM=Llama 3.1-8B, rank=322026.056.23
AdaLoRALLM=Llama 3.1-8B2026.056.19
rsLoRALLM=Llama 3.1-8B2026.056.18
LoRA-Over-MPO_PLLM=Llama 3.1-8B2026.056.16
LoRALLM=Llama 3.1-8B2026.056.15
OLoRALLM=Llama 3.1-8B2026.056.13
LoRA-Over-MPOLLM=Llama 3.1-8B2026.056.12
PiSSALLM=Llama 3.1-8B2026.056.08
LoRA-Over-SVDLLM=Llama 3.1-8B2026.056.01
LoRA-GALLM=Llama 3.1-8B2026.055.99
DoRALLM=Llama 2-7B2026.055.97