OpenAI has made a significant shift back to its open-source roots by releasing two new open-weight language models: gpt-oss-120b and gpt-oss-20b. This marks the first public release of AI model weights since GPT-2 in 2019. These models allow users to download, examine, and fine-tune them without relying on cloud APIs, though training data remains undisclosed due to legal concerns. The gpt-oss-120b model is designed for high-performance systems, while the gpt-oss-20b is compact enough for standard laptops. Both models are licensed under Apache 2.0, promoting ease of use and monetization for developers. Despite early tests indicating a tendency for hallucinations, they demonstrate robust reasoning capabilities via a mixture-of-experts architecture. This move aims to lower barriers for emerging markets and smaller organizations as OpenAI acknowledges its previous reluctance to embrace open-source principles, a change prompted by competition from models like DeepSeek.
Source link
Share
Read more