OpenAI releases GPT OSS: New open-source model family

AI Impact Summary

OpenAI is releasing a new family of open-source models, GPT OSS, with two models (117B and 21B parameters) designed for reasoning and agentic tasks. These models utilize a Mixture-of-Experts (MoE) architecture with 4-bit quantization (MXFP4) to enable fast inference and low resource usage, making them suitable for a wide range of applications, including local deployments and consumer hardware. The release includes support for various inference frameworks like vLLM, llama.cpp, and Hugging Face’s Inference Providers.