Particle.news
Download on the App Store
48 ARTICLES
·
3w ago

OpenAI Releases Its First Open-Weight Models Since GPT-2

This move reflects a strategic shift to democratize AI access under a permissive Apache 2.0 license.

Image
Image
PCs powered by an RTX 5090 GPU can get performance of up to 256 tokens per second when running OpenAI's new models.

Overview

  • OpenAI has launched gpt-oss-120b and gpt-oss-20b under the Apache 2.0 license, marking its first open-weight releases since GPT-2 and offering free downloads via Hugging Face and GitHub.
  • The larger gpt-oss-120b runs on a single Nvidia GPU while the slimmer gpt-oss-20b can operate on consumer devices with 16 GB of RAM, enabling offline deployment from laptops to workstations.
  • Both models support advanced chain-of-thought reasoning and can integrate tools such as web search and Python code execution to power multi-step AI workflows.
  • Safety evaluations by OpenAI and third-party experts found that neither model reaches the company’s “high capability” threshold for biological or chemical risks, even after fine-tuning.
  • CEO Sam Altman described the releases as a return to OpenAI’s founding mission of ensuring AGI benefits all humanity and fostering an open AI ecosystem in the United States.