AI BriefingOpenAI Releases Open Source Model 'GPT OSS'
Key point
OpenAI has released the GPT OSS series, an open-weight model lineup under the Apache 2.0 license.
Details
OpenAI has unveiled GPT OSS, a family of open-weight models designed for powerful reasoning and agentic tasks. The new models adopt the Apache 2.0 license, allowing developers to use them more freely.
The models are offered in two sizes:
- gpt-oss-120b: A 117B parameter model that can run on a single H100 GPU.
- gpt-oss-20b: A 21B parameter model that operates within 16GB of memory, optimized for consumer hardware and on-device environments.
Both models are based on a Mixture-of-Experts (MoE) architecture and use MXFP4 4-bit quantization to increase inference speed and reduce resource usage.
Key technical features include:
- 128K context window support (using RoPE)
- SwiGLU activation function and Token-choice MoE architecture
- Support for Chain-of-Thought and adjustable reasoning levels
- Same tokenizer as GPT-4o
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.