List your company
ResearchFormerly WizWand

General Reasoning on ARC-C

98AccuracyMAE
15.00836.55458.179.646Jul 2, 2025Aug 25, 2025Oct 19, 2025Dec 13, 2025Feb 6, 2026Apr 2, 2026May 27, 2026
Updated 5mo ago

Evaluation results

MethodLinks
MAEModel=Llama-3.2-3B-Instruct2026.0598——
COSEModel=Qwen3-4B2026.0597.4——
COSEModel=Qwen3-0.6B2026.0597——
BaseModel=Qwen3-4B2026.0597——
MAEModel=Qwen3-4B2026.0596.9——
COSEModel=Llama-3.2-3B-Instruct2026.0596.8——
AZRModel=Qwen3-4B2026.0596.8——
R-ZeroModel=Qwen3-4B2026.0596.7——
Full-step decodingModel=Dream-7B-Instruct, Decoding Strategy=Full-step, Evaluation Setting=Zero-shot2025.0888.1——
ProphetModel=Dream-7B-Instruct, Decoding Strategy=Prophet, Evaluation Setting=Zero-shot2025.0887.9-0.22.61
MAEModel=Qwen2.5-3B-Instruct2026.0584.2——
Prefix-RFTBase Model=Qwen2.5-Math-7B2025.0784——
ProphetModel=LLaDA-8B-Instruct, Decoding Strategy=Prophet, Evaluation Setting=Zero-shot2025.0883.50.31.88
UFTBase Model=Qwen2.5-Math-7B2025.0783.4——
Full-step decodingModel=LLaDA-8B-Instruct, Decoding Strategy=Full-step, Evaluation Setting=Zero-shot2025.0883.2——
COSEModel=Qwen2.5-3B-Instruct2026.0583——
AZRModel=Qwen2.5-3B-Instruct2026.0582.8——
RFTBase Model=Qwen2.5-Math-7B2025.0782.3——
R-ZeroModel=Qwen2.5-3B-Instruct2026.0581.9——
BaseModel=Qwen2.5-3B-Instruct2026.0580.6——
LUFFYBase Model=Qwen2.5-Math-7B2025.0780.5——
ReLIFTBase Model=Qwen2.5-Math-7B2025.0776.2——
SFTBase Model=Qwen2.5-Math-7B2025.0775.2——
AZRModel=Llama-3.2-3B-Instruct2026.0573.4——
R-ZeroModel=Llama-3.2-3B-Instruct2026.0573.2——
SFT+RFTBase Model=Qwen2.5-Math-7B2025.0772.4——
MAEModel=Qwen3-0.6B2026.0572.3——
R-ZeroModel=Qwen3-0.6B2026.0571.4——
RL w/ SFT LossBase Model=Qwen2.5-Math-7B2025.0771.2——
Oat-ZeroBase Model=Qwen2.5-Math-7B2025.0770.1——
AZRModel=Qwen3-0.6B2026.0569.8——
BaseModel=Qwen3-0.6B2026.0568——
BaseModel=Llama-3.2-3B-Instruct2026.0567——
SimpleRL-ZeroBase Model=Qwen2.5-Math-7B2025.0730.2——
Qwen2.5-Math-7B2025.0718.2——