gpt-oss-20b
An Apache 2.0 licensed open-weight 21B-parameter mixture-of-experts language model from OpenAI optimized for edge and on-device deployments.
- 21B total parameters with 3.6B active parameters via Mixture-of-Experts (MoE) architecture
- Apache 2.0 open-source license
- 131,000-token context window
- Optimized for edge and local deployment with a 16GB memory requirement
Identity
gpt-oss-20b is an open-weight language model developed by [[OpenAI]], released on August 5, 2025 (model profile). It is an Apache 2.0 licensed model utilizing a Mixture-of-Experts (MoE) architecture (model profile). While the model profile identifies it as a 21-billion parameter model, the model name and subsequent reporting in 2026 refer to it as a 20-billion parameter model (Signal + Noise wire item).
What it is
The model is a general-purpose language model optimized for edge and on-device deployments (model profile). It utilizes a Mixture-of-Experts design with 21 billion total parameters, of which 3.6 billion are active during inference (model profile). It supports a context window of 131,072 tokens and is designed to operate within a 16GB memory footprint (model profile).
Capabilities & benchmarks
As reported in the model profile, gpt-oss-20b features a 13