gpt-oss
Two open-weight language models from OpenAI, gpt-oss-120b and gpt-oss-20b, designed for reasoning and agentic tasks such as function calling, web browsing and code execution, released under the Apache 2.0 license.
gpt-oss is a series of open-weight language models from OpenAI, released as gpt-oss-120b, a 117-billion-parameter model with 5.1 billion active parameters that fits on a single 80GB GPU, and gpt-oss-20b, a 21-billion-parameter model with 3.6 billion active parameters for lower-latency or local use, running within 16GB of memory after MXFP4 quantization of its mixture-of-experts weights.
Both models use OpenAI's harmony response format and expose configurable reasoning effort, low, medium or high, along with full chain-of-thought output. They support function calling, web browsing, Python code execution and structured outputs, and can be fine-tuned for a specific use case. They can be run through the Transformers library, a reference PyTorch or Triton implementation, or a Metal implementation.