List your company
ResearchFormerly WizWand

General Reasoning on ARC Challenge (Score)

94.58ARC ScoreInfiGFusion
64.607272.388680.1787.9514May 20, 2025Jul 19, 2025Sep 17, 2025Nov 16, 2025Jan 15, 2026Mar 16, 2026May 16, 2026
Updated 5mo ago

Evaluation results

MethodLinks
InfiGFusionModel Size=14B, GPU Hour=1952025.0594.58
SFT-DPOModel Size=14B, GPU Hours=502025.0594.58
InfiFusionModel Size=14B, GPU Hour=1602025.0594.24
InfiFusion*Model Size=14B, GPU Hours=160, Subset Source Models=true2025.0594.24
SFT-IPOModel Size=14B, GPU Hours=502025.0594.24
SFT-WRPOModel Size=14B, GPU Hours=572025.0594.24
InfiFPO*Model Size=14B, GPU Hours=55, Subset Source Models=true2025.0594.24
InfiFPOModel Size=14B, GPU Hours=582025.0594.24
Phi-4Model Size=14B, GPU Hour=∼1.0M2025.0593.9
Phi-4Model Size=14B, GPU Hours=∼1.0M2025.0593.9
Pivot-SFTModel Size=14B, GPU Hour=1202025.0593.56
FuseChatModel Size=14B, GPU Hour=6502025.0593.56
FuseChat*Model Size=14B, GPU Hours=650, Subset Source Models=true2025.0593.56
SFTModel Size=14B, GPU Hours=152025.0593.56
Qwen2.5-InstructModel Size=14B, GPU Hour=∼1.8M2025.0592.2
Qwen2.5-InstructModel Size=14B, GPU Hours=∼1.8M2025.0592.2
MiniLogitModel Size=14B, GPU Hour=2202025.0592.19
FuseLLMModel Size=14B, GPU Hour=2252025.0592.08
FuseLLM*Model Size=14B, GPU Hours=225, Subset Source Models=true2025.0592.08
Mistral-SmallModel Size=24B, GPU Hour=∼1.6M2025.0591.86
Mistral-SmallModel Size=24B, GPU Hours=∼1.6M2025.0591.86
Qwen2.5-CoderModel Size=14B, GPU Hour=∼1.8M2025.0589.49
Qwen2.5-CoderModel Size=14B, GPU Hours=∼1.8M2025.0589.49
Task ArithmeticModel Scale=Llama-3.1-8B, Quantization Bit-width=Full-precision, Merging Algorithm=Task Arithmetic2026.0577.99
AWQModel Scale=Llama-3.1-8B, Quantization Bit-width=4-bit, Merging Algorithm=Task Arithmetic2026.0576.62
GPTQModel Scale=Llama-3.1-8B, Quantization Bit-width=4-bit, Merging Algorithm=Task Arithmetic2026.0576.45
E-PMQModel Scale=Llama-3.1-8B, Quantization Bit-width=4-bit, Merging Algorithm=Task Arithmetic2026.0576.28
Task ArithmeticModel Scale=Llama-3.1-3B, Quantization Bit-width=Full-precision, Merging Algorithm=Task Arithmetic2026.0572.7
AWQModel Scale=Llama-3.1-3B, Quantization Bit-width=4-bit, Merging Algorithm=Task Arithmetic2026.0572.44
Gemma-3-InstructModel Size=12B2025.0571.19
E-PMQModel Scale=Llama-3.1-3B, Quantization Bit-width=4-bit, Merging Algorithm=Task Arithmetic2026.0570.99
GPTQModel Scale=Llama-3.1-3B, Quantization Bit-width=4-bit, Merging Algorithm=Task Arithmetic2026.0570.9
Qwen2.5-MathModel Size=7B, GPU Hours=∼0.5M2025.0565.76