List your company
ResearchFormerly WizWand

Dialogue on MT-Bench

9.3MT-Bench ScoreGPT-4o
5.06726.16617.2658.3639Feb 24, 2025May 16, 2025Aug 6, 2025Oct 27, 2025Jan 17, 2026Apr 9, 2026Jun 30, 2026
Updated 4mo ago

Evaluation results

MethodLinks
GPT-4oEvaluation Protocol=Closed-Source API, Emission (gCO2/q)=4.52†, Throughput (Tok/s)=55.22026.039.3
Claude 3.5 SonnetEvaluation Protocol=Closed-Source API, Emission (gCO2/q)=3.85†, Throughput (Tok/s)=62.12026.039.1
Gemini 2.5 ProEvaluation Protocol=Closed-Source API, Emission (gCO2/q)=3.90†, Throughput (Tok/s)=58.42026.039
EcoThinkEvaluation Protocol=Adaptive Inference, Emission (gCO2/q)=1.32, Throughput (Tok/s)=148.62026.038.9
Qwen-3-8B-InstructEvaluation Protocol=Standard CoT, Emission (gCO2/q)=2.12, Throughput (Tok/s)=98.52026.038.6
FrugalGPT (Cascade)Evaluation Protocol=Standard CoT, Emission (gCO2/q)=1.95, Throughput (Tok/s)=88.52026.038.5
Llama-3.1-8B-InstructEvaluation Protocol=Standard CoT, Emission (gCO2/q)=2.15, Throughput (Tok/s)=95.82026.038.2
SPARDBackbone=Qwen2.5-7B-Instruct2026.048.138
+ GRPOavgBackbone=Qwen2.5-7B-Instruct2026.047.975
+ GRPOimpBackbone=Qwen2.5-7B-Instruct2026.047.862
+ DPOBackbone=Qwen2.5-7B-Instruct2026.047.837
SPARDBackbone=Qwen3-8B2026.047.831
+ GRPOrmBackbone=Qwen2.5-7B-Instruct2026.047.812
BaseBackbone=Qwen2.5-7B-Instruct2026.047.793
+ DPOBackbone=Qwen3-8B2026.047.762
+ SFTBackbone=Qwen2.5-7B-Instruct2026.047.756
+ GRPOimpBackbone=Qwen3-8B2026.047.725
+ GRPOrmBackbone=Qwen3-8B2026.047.681
+ GRPOavgBackbone=Qwen3-8B2026.047.681
BaseBackbone=Qwen3-8B2026.047.506
+ SFTBackbone=Qwen3-8B2026.047.162
NA-LoRABackbone=Llama-3.1-8B-Base, Rank (r)=82026.066.39
DoRABackbone=Llama-3.1-8B-Base2026.066.32
GoRABackbone=Llama-3.1-8B-Base2026.066.3
LoRA+Backbone=Llama-3.1-8B-Base2026.066.27
Bi-LoRACost=×1, Backbone=Llama 2-7B, Evaluation Protocol=Fine-tuning2025.086.26
LoRA-ProBackbone=Llama-3.1-8B-Base2026.066.21
NA-LoRABackbone=Llama-3.1-8B-Base, Rank (r)=322026.066.2
AdaLoRABackbone=Llama-3.1-8B-Base2026.066.18
Full FTCost=×1, Backbone=Llama 2-7B, Evaluation Protocol=Fine-tuning2025.086.16
rsLoRABackbone=Llama-3.1-8B-Base2026.066.16
NA-LoRABackbone=Llama-3.1-8B-Base, Rank (r)=1282026.066.12
LoRA-nBARCost=×1, Backbone=Llama 2-7B, Evaluation Protocol=Fine-tuning2025.086.1
PiSSABackbone=Llama-3.1-8B-Base2026.066.1
LoRABackbone=Llama-3.1-8B-Base2026.066.08
ROPCost=×1, Backbone=Llama 2-7B, Evaluation Protocol=Fine-tuning2025.086.05
GOATBackbone=Llama-2-7B2025.026.01
RWP_fullCost=×1, Backbone=Llama 2-7B, Evaluation Protocol=Fine-tuning2025.086.01
LoRA-GABackbone=Llama-3.1-8B-Base2026.066
Flat-LoRACost=×1, Backbone=Llama 2-7B, Evaluation Protocol=Fine-tuning2025.085.98
DoRABackbone=Llama-2-7B2025.025.97
LoRA-SAMCost=×2, Backbone=Llama 2-7B, Evaluation Protocol=Fine-tuning2025.085.97
LoRA-oBARCost=×1, Backbone=Llama 2-7B, Evaluation Protocol=Fine-tuning2025.085.97
DoRABackbone=Llama-2-7B-Base2026.065.97
LoRA-GABackbone=Llama-2-7B-Base2026.065.95
LoRA-LookSAMCost=×1.x, Backbone=Llama 2-7B, Evaluation Protocol=Fine-tuning2025.085.94
LoRACost=×1, Backbone=Llama 2-7B, Evaluation Protocol=Fine-tuning2025.085.92
NA-LoRABackbone=Llama-2-7B-Base, Rank (r)=82026.065.92
FullBackbone=Llama-3.1-8B-Base2026.065.88
MoLoRABackbone=Llama-2-7B2025.025.84
HydraLoRABackbone=Llama-2-7B2025.025.82
RWP_LoRACost=×1, Backbone=Llama 2-7B, Evaluation Protocol=Fine-tuning2025.085.8
LoRA-ESAMCost=×1.x, Backbone=Llama 2-7B, Evaluation Protocol=Fine-tuning2025.085.8
NA-LoRABackbone=Llama-2-7B-Base, Rank (r)=1282026.065.78
LoRA-ProBackbone=Llama-2-7B-Base2026.065.72
NA-LoRABackbone=Llama-2-7B-Base, Rank (r)=322026.065.72
LoRA+Backbone=Llama-2-7B-Base2026.065.71
LoRABackbone=Llama-2-7B2025.025.61
LoRABackbone=Llama-2-7B-Base2026.065.61
GoRABackbone=Llama-2-7B-Base2026.065.61
AdaLoRABackbone=Llama-2-7B-Base2026.065.57
Full FTBackbone=Llama-2-7B2025.025.56
PiSSABackbone=Llama-2-7B2025.025.3
FullBackbone=Llama-2-7B-Base2026.065.3
PiSSABackbone=Llama-2-7B-Base2026.065.3
rsLoRABackbone=Llama-2-7B-Base2026.065.25
MiLoRABackbone=Llama-2-7B2025.025.23