gpt-oss-20b
OpenAI · 2025-08-05 · 20.9B parameters
gpt-oss-20b is OpenAI's smaller open-weight reasoning model in the Apache 2.0-licensed gpt-oss family, sized for lower-latency, local, or specialized use cases. Its checkpoint contains exactly 20,914,757,184 logical parameters after unpacking MXFP4 weights and excluding quantization scales; OpenAI reports approximately 3.6B active parameters per token. It is a text-only mixture-of-experts model with configurable reasoning effort and tool-use support.