gpt-oss 120B

gpt-oss 120B is OpenAI's larger open-weight model — a 117B parameter Mixture-of-Experts with 5.1B active parameters per token, released under Apache 2.0. Like its 20B sibling it ships natively in MXFP4 quantization, keeping the full model to a ~65 GB footprint that fits on a single 80 GB datacenter GPU. Performance lands near o4-mini on core reasoning benchmarks, with strong tool use, browsing, and adjustable reasoning effort. For local use it's a Mac Studio / multi-GPU proposition: 96 GB+ unified memory runs it comfortably, while 64 GB configurations are out of reach without offloading.