New powerful (GPT based) model from Microsoft, Small, fast and functional all in one just because of the new paper. Developer: Microsoft Research, Machine Lear

New powerful (GPT based) model from Microsoft, Small, fast and functional all in one just because of the new paper. Developer: Microsoft Research, Machine Learning and Optimization (MLO) Group Model Architecture: Mixture-of-Experts (MoE) variant of the transformer architecture (gpt-oss family). Parameters: 20 Billion (3.6B activated) Inputs: Natural language optimization problem description. Context Length: 128,000 tokens Paper for the method used: https://arxiv.org/pdf/2509.22979 https://arxiv.org/pdf/2509.22979