gpt-oss 20B

gpt-oss 20B is OpenAI's smaller open-weight model — a 21B parameter Mixture-of-Experts with only 3.6B active parameters per token. It ships natively in MXFP4 quantization, which is how it was post-trained, so the ~14 GB download is effectively the full-quality model rather than a lossy compression of it. It delivers o3-mini-class reasoning with adjustable reasoning effort (low/medium/high) and strong agentic tool use, making it the go-to answer for capable local reasoning on 16 GB cards and 16-24 GB Macs. One of the most-pulled models on Ollama since its release.