OpenAI has released two new open-weight AI models, GPT OSS-120b and GPT OSS-20b, under the Apache 2.0 license. These models, comparable to o4-mini, are designed for various applications and are optimized for reasoning tasks.
OpenAI has launched two new open-weight AI models, marking its first open release since GPT-2 more than five years ago. The newly unveiled models, GPT OSS-120b and GPT OSS-20b, are now available for download via Hugging Face under the permissive Apache 2.0 licence, making them freely accessible for developers and enterprises alike.
Scroll to load tweet…
Announcing the new models via an X post, CEO Sam Altman said “gpt-oss is out! We made an open model that performs at the level of o4-mini and runs on a high-end laptop (WTF!!) (and a smaller one that runs on a phone). super proud of the team; big triumph of technology.”
Scroll to load tweet…
OpenAI’s gpt-oss-120b and gpt-oss-20b models are available under the Apache 2.0 license. While the smaller gpt-oss-20b is intended for usage on devices with limited hardware, the bigger variant, gpt-oss-120b, is said to equal the performance of OpenAI’s proprietary o4-mini model. Both models may be used to create chatbots, help with coding, and other AI-powered applications, and they are optimised for reasoning tasks. This coincides with OpenAI’s planned release of GPT-5.
Here’s what we know so far
According to OpenAI, the gpt-oss-120b model can operate on a single 80GB GPU, whilst the gpt-oss-20b variant is made for machines with a mere 16GB of RAM. According to OpenAI, both models can utilise tools, call functions, and reason. In order to balance performance and speed, they might also modify their reasoning efforts based on the job. According to the corporation, it has used its own safety standards to undertake safety tests and training for these models.
Scroll to load tweet…
The models were trained on extensive text datasets that concentrated on general knowledge, STEM disciplines, and coding. To test the models for applications like secure on-premise installations, OpenAI collaborated with early partners including Orange and AI Sweden.A Transformer architecture with cutting-edge memory-efficiency algorithms is used in both versions. There are 117 billion parameters in the bigger model and 21 billion in the smaller one.