TLDR: OpenAI has released two new open-weight large language models, GPT-OSS 120B and GPT-OSS 20B, marking a significant shift towards broader accessibility for developers. These models, available under the Apache 2.0 license, offer strong performance and cost-effectiveness, with the larger model matching OpenAI’s o4-mini and the smaller optimized for mobile devices. This move also sees OpenAI’s models available on Amazon Web Services (AWS) for the first time, breaking Microsoft’s prior exclusive cloud distribution for OpenAI’s offerings.
On August 6, 2025, OpenAI announced the release of its new open-weight large language models, GPT-OSS 120B and GPT-OSS 20B, signaling a strategic evolution in the company’s approach to model distribution and accessibility. This development, initially highlighted by The Information reporting on OpenAI’s ‘secret strategy’ for its open models, represents OpenAI’s first open model release since GPT-2 in 2019.
The GPT-OSS models are designed to empower developers with more cost-effective and flexible AI solutions. The GPT-OSS 120B model, with 117 billion parameters, is reported to match the performance of OpenAI’s o4-mini and can operate efficiently on a single 80GB GPU. The smaller GPT-OSS 20B model, with 21 billion parameters, is optimized for devices with limited memory, such as those with 16GB, making it suitable for high-end laptops and even phones.
These models are released under the Apache 2.0 license, granting developers broad rights to use, modify, and redistribute them. While termed ‘open-weight,’ it’s important to note that OpenAI has not provided access to the underlying training code or datasets, maintaining a balance between accessibility and safeguarding proprietary research. Both models are equipped with advanced capabilities including reasoning, function calling, tool use, and the ability to adjust reasoning efforts to balance speed and performance. They also support instruction following, Python code execution, and web search functionalities.
OpenAI emphasized its commitment to safety, stating that it conducted extensive safety training and evaluations for these models, including adversarial fine-tuning, to ensure compliance with its Preparedness Framework for safe AI deployment. Furthermore, a red teaming challenge was launched for the gpt-oss-20b model, inviting participants to identify and report vulnerabilities, including reward hacking, deception, and hidden motivations.
Also Read:
- OpenAI’s Advanced Open-Weight Models Now Available on Amazon Web Services
- OpenAI and NVIDIA Unveil New Open-Weight AI Models for Global Inference Infrastructure
A significant aspect of this release is the expanded cloud availability. For the first time, OpenAI’s open-weight models are being offered on Amazon Web Services (AWS), marking a departure from Microsoft’s previous exclusive hold on cloud distribution for OpenAI’s models. While OpenAI’s proprietary models remain exclusive to Microsoft Azure, the availability of GPT-OSS on AWS is a strategic move that could intensify competition in the cloud AI market. AWS has highlighted the new models’ competitive price performance compared to offerings from Gemini, DeepSeek, and even OpenAI’s own o4 model. OpenAI CEO Sam Altman expressed pride in the achievement, stating via an X post, ‘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.’ This release underscores OpenAI’s evolving strategy to balance cutting-edge proprietary research with broader community access and innovation.
