OpenAI has released two open-weight language models, the gpt-oss-120b and gpt-oss-20b, allowing high performance on consumer hardware. The 120B model requires a GPU with 80GB memory, while the smaller 20B version operates on 16GB, making it more accessible. Both models leverage a mixture-of-experts approach, offering performance comparable to OpenAI's commercial offerings. Available under the Apache 2.0 license, they enable unrestricted use, modification, and potential commercialization. The 120B version excelled in reasoning benchmarks, surpassing existing models, while the 20B version matched or exceeded its predecessor across several metrics. OpenAI has emphasized safety, filtering harmful data and fine-tuning the models to prevent unsafe responses. The release coincides with speculation about GPT-5's future, indicating the competitive landscape of AI continues to evolve rapidly, with OpenAI asserting a commitment to responsible AI development and safety testing.

Source 🔗