OpenAI introduced the gpt-oss-120b and gpt-oss-20b models, which are open weight large language models designed to run on smaller hardware. The gpt-oss-120b model features 117 billion parameters and can operate on a single A100 GPU, while the gpt-oss-20b has 21 billion parameters and can run with just 16GB of memory. Both models employ a mixture of experts approach for efficiency, performing comparably to the o4-mini model in various benchmarks. These models are intended to enhance transparency and safety in AI technology.
OpenAI has released two open weight LLMs, gpt-oss-120b and gpt-oss-20b, which can perform similarly to recent small language models on accessible hardware.
Both models utilize a mixture of experts approach, allowing for faster inference and less costly pre-training through selective activation of parameters.
The larger model, gpt-oss-120b, has 117 billion parameters and outperforms o4-mini in health and expert questioning benchmarks, though lags slightly in code completion.
Sam Altman stated that gpt-oss is a state-of-the-art open-weights reasoning model, emphasizing its strong real-world performance and usability on local systems.
Collection
[
|
...
]