ArchitectureTrainingSafetyEvaluation
gpt-oss-120b & gpt-oss-20b Model Card
OpenAI·August 11, 2025
OpenAI
View on arXivTL;DR
OpenAI’s first open-weight LLMs since GPT-2: two MoE reasoning models (116.8B and 20.9B) trained largely via distillation, with MXFP4 quantization letting the large model fit a single 80GB GPU. Apache 2.0.
Why it matters
Marked OpenAI’s return to open weights; the 20B model runs on consumer hardware, broadening access to o4-mini-class reasoning.