Photo by Anastasiya Gepp on Pexels

OpenAI has launched ‘gpt-oss,’ its first open-weight large language models (LLMs) since 2019, signaling a renewed focus on open-source AI. Available in two sizes, these models offer performance on par with OpenAI’s previously available o3-mini and o4-mini, but with the crucial difference that they are freely downloadable, runnable, and modifiable.

This move positions OpenAI to compete directly with the growing influence of open-source LLMs, particularly those emerging from China, such as DeepSeek and Alibaba’s Qwen series. According to OpenAI’s Casey Dvorak, a significant portion of enterprise and startup clients are already leveraging open models, creating a clear market for OpenAI to address.

The ‘gpt-oss’ models are released under the permissive Apache 2.0 license, enabling commercial use and fostering broader adoption. This licensing approach, lauded by experts like Nathan Lambert of the Allen Institute for AI as a step forward, allows for more freedom compared to alternatives.

Beyond competition, this release serves as a strategic move for OpenAI to maintain its leadership in AI research. By making the models accessible for in-depth analysis and manipulation, OpenAI hopes to foster innovation that can be incorporated into future developments. The timing also underscores geopolitical considerations, as OpenAI aims to address concerns surrounding content restrictions and align with U.S. AI priorities, particularly regarding computational infrastructure. Ultimately, ‘gpt-oss’ represents a significant stride towards balanced global AI development and the fortification of OpenAI’s position in the rapidly evolving technological landscape.