OpenAI and NVIDIA have launched two revolutionary open-weight large language models (LLMs) — gpt-oss-120b and gpt-oss-20b. Designed for developers, researchers, startups, and enterprises, these models enhance reasoning capabilities and offer exceptional performance and flexibility for various deployment environments. Trained on NVIDIA’s H100 GPUs, they achieve impressive inference speeds of 1.5 million tokens per second, especially on Blackwell-powered GB200 NVL72 systems. Both models are released under the Apache 2.0 license, supporting commercial and research usage. The gpt-oss-120b model rivals OpenAI’s o4-mini, while the gpt-oss-20b is optimized for edge devices. With compatibility across leading frameworks like Hugging Face and TensorRT-LLM, developers enjoy the freedom to leverage preferred tools. Safety measures, including independent reviews and adversarial testing, ensure reliable performance. By fostering open-source AI, OpenAI and NVIDIA empower the next generation of innovators in the tech landscape.
Source link
Share
Read more