OpenAI has made a major shift in strategy, releasing its first open‑weight AI models since GPT‑2 in 2019. The two models—gpt‑oss‑120B and gpt‑oss‑20B—are now downloadable under the Apache 2.0 license, allowing users to run, inspect, modify, and redistribute them freely even on personal devices like laptops and high‑end smartphones. 

These text‑only models support advanced chain‑of‑thought reasoning and autonomous agent workflows, such as web browsing, code execution, and software navigation. The gpt‑oss‑20B variant runs comfortably on machines with 16 GB RAM, while gpt‑oss‑120B delivers high-end performance comparable to OpenAI's proprietary o4‑mini.

OpenAI emphasizes that this release complements—rather than replaces—its closed models and is guided by a rigorous safety evaluation framework aimed at minimizing misuse, including bioweapon creation risks.

This pivot toward openness responds to competitive pressure from rivals like DeepSeek and Meta, and signals OpenAI’s renewed commitment to democratizing AI access and strengthening developer ecosystems.