List your company
ResearchFormerly WizWand

General Instruction Following on Arena-Hard

22.1ScoreTAG-INSTRUCT
-0.640645.2631811.16717.07082Apr 11, 2025Jun 4, 2025Jul 28, 2025Sep 20, 2025Nov 13, 2025Jan 6, 2026Mar 2, 2026
Updated 8mo ago

Evaluation results

MethodLinks
TAG-INSTRUCTBase Model=LLaMA3-8b, Teacher Model=Ministral-Instruct-8b, #Convs=5K2025.0522.1—
CodecLMBase Model=LLaMA3-8b, Teacher Model=Ministral-Instruct-8b, #Convs=5K2025.0519.2—
Evol-InstructBase Model=LLaMA3-8b, Teacher Model=Ministral-Instruct-8b, #Convs=5K2025.0516.8—
Auto-Instruct-EvolBase Model=LLaMA3-8b, Teacher Model=Ministral-Instruct-8b, #Convs=5K2025.0516.8—
Tree-InstructBase Model=LLaMA3-8b, Teacher Model=Ministral-Instruct-8b, #Convs=5K2025.0516.6—
TAG-INSTRUCTBase Model=LLaMA3.2-3b, Teacher Model=Ministral-Instruct-8b, #Convs=5K2025.0512.9—
Evol-InstructBase Model=LLaMA3.2-3b, Teacher Model=Ministral-Instruct-8b, #Convs=5K2025.0512.7—
CodecLMBase Model=LLaMA3.2-3b, Teacher Model=Ministral-Instruct-8b, #Convs=5K2025.0512.7—
Auto-Instruct-EvolBase Model=LLaMA3.2-3b, Teacher Model=Ministral-Instruct-8b, #Convs=5K2025.0510.8—
Tree-InstructBase Model=LLaMA3.2-3b, Teacher Model=Ministral-Instruct-8b, #Convs=5K2025.0510.4—
WizardLM-dataBase Model=LLaMA3-8b, Teacher Model=Ministral-Instruct-8b, #Convs=192K2025.054.5—
WizardLM-dataBase Model=LLaMA3.2-3b, Teacher Model=Ministral-Instruct-8b, #Convs=192K2025.053.7—
Alpaca-5kBase Model=LLaMA3-8b, Teacher Model=Ministral-Instruct-8b, #Convs=5K2025.053.1—
Alpaca-CleanBase Model=LLaMA3-8b, Teacher Model=Ministral-Instruct-8b, #Convs=52K2025.053—
Alpaca-CleanBase Model=LLaMA3.2-3b, Teacher Model=Ministral-Instruct-8b, #Convs=52K2025.052.8—
Alpaca-5kBase Model=LLaMA3.2-3b, Teacher Model=Ministral-Instruct-8b, #Convs=5K2025.052.6—
Qwen3-235B-A22B-R-TAPArchitecture=MoE, Parameter Count=235B2026.030.971—
Gemini2.5-Pro2026.030.964—
Qwen3-235B-A22BArchitecture=MoE, Parameter Count=235B2026.030.956—
Qwen3-32B-R-TAPArchitecture=Dense, Parameter Count=32B2026.030.956—
AM-Thinking-v1-R-TAPArchitecture=Dense, Parameter Count=32B2026.030.943—
Qwen3-32BArchitecture=Dense, Parameter Count=32B2026.030.938—
DeepSeek-R1Architecture=MoE, Parameter Count=671B2026.030.932—
AM-Thinking-v1Architecture=Dense, Parameter Count=32B2026.030.925—
OpenAI-o1 (2024-12-17)2026.030.921—
OpenAI-o3-mini (Medium)Configuration=Medium2026.030.89—
Nemetron-Ultra-253BArchitecture=Dense, Parameter Count=256B2026.030.87—
RefEvalBase Model=Qwen2.5-7B-SFT, Training Method=DPO from DSV3-Distill2026.020.741—
ArmoRMBase Model=Qwen2.5-7B-SFT, Training Method=DPO from DSV3-Distill2026.020.722—
RefFreeBase Model=Qwen2.5-7B-SFT, Training Method=DPO from DSV3-Distill2026.020.718—
ROUGEBase Model=Qwen2.5-7B-SFT, Training Method=DPO from DSV3-Distill2026.020.674—
BERTScoreBase Model=Qwen2.5-7B-SFT, Training Method=DPO from DSV3-Distill2026.020.645—
RefEvalBase Model=Llama-3-8B-Instruct, Training Method=DPO from DSV3-Distill2026.020.587—
ArmoRMBase Model=Llama-3-8B-Instruct, Training Method=DPO from DSV3-Distill2026.020.586—
ArmoRM-BaseBase Model=Qwen2.5-7B-SFT2026.020.586—
DSV3-DistillBase Model=Qwen2.5-7B-SFT2026.020.565—
RefFreeBase Model=Llama-3-8B-Instruct, Training Method=DPO from DSV3-Distill2026.020.538—
BERTScoreBase Model=Llama-3-8B-Instruct, Training Method=DPO from DSV3-Distill2026.020.53—
ROUGEBase Model=Llama-3-8B-Instruct, Training Method=DPO from DSV3-Distill2026.020.521—
GeniusTraining Data Source=Magpie, Training Sample Size=25K2025.040.5—
GeniusTraining Data Source=OpenHermes, Training Sample Size=32K2025.040.5—
DSV3-DistillBase Model=Llama-3-8B-Instruct2026.020.422—
ArmoRM-BaseBase Model=Llama-3-8B-Instruct2026.020.404—
LLaMA3.1-8B-Instruct2025.040.3031—
BaseBase Model=Llama-3-8B-Instruct2026.020.271—
BaseBase Model=Qwen2.5-7B-SFT2026.020.234—