OpenAI launches two new models for free commercial use that can run on a laptop
OpenAI has announced GPT-OSS, a new open-weight model family available for free download, customization, and commercial use under the Apache 2.0 license. The launch includes two variants: a 120-billion-parameter model that performs comparably to OpenAI's o4-mini, and a 20 billion-parameter model that matches o3-mini and runs efficiently on local hardware with only 16GB of VRAM. The larger model can operate on a single Nvidia GPU and includes an adversarially fine-tuned variant for improved safety evaluation.
Both models support few-shot function calling, chain-of-thought reasoning, and HealthBench tasks, and are compatible with OpenAI's Responses API. They are natively quantized in MXFP4 for efficient local memory use and are available through platforms like Hugging Face, Databricks, Microsoft Azure, and Amazon Web Services. Microsoft has also released a GPU-optimized GPT-OSS-20B model for Windows PCs via Foundry Local and the AI Toolkit for Visual Studio Code.
While developers can experiment with GPT-OSS via OpenAI's online playground, OpenAI has open-sourced tools including a harmony renderer in Python and Rust to support adoption, along with reference implementations for PyTorch and Apple’s Metal
- AI Chatbot
- Freemium • Proprietary
Related news
External links
- Introducing gpt-ossOpenAI Blog • Official source
- OpenAI finally releases its open-weight models optimized for laptops and smartphonesNeowin
- OpenAI launches two 'open' AI reasoning modelsTechCrunch
- OpenAI releases a free GPT model that can run on your laptopThe Verge
Comments
OpenAI misses one thing: the infrastructure (i.e. big computing centers to compute always bigger models). With Microsoft that is less and less willing to continue burning "money" (Azure credits) for OpenAI, Softbank that is still supposed to build the Stargate computing center (and own the trademark) that has not be started yet, and Nvidia selling even more expensive GPUs, OpenAI starts to fell alone. So, another genius idea from OpenAI is to waste its billions ($10B from Softbank, supposed another $10B at the end of the year, plus $8.3B recently from investors, plus the some billions in cash) into building other computing centers, exactly what Amazon, Meta, Google, Microsoft and Oracle are doing by burning a giant amount of money doing so (and earning less than 10% in return). OpenAI is just inflating a big bubble to build its empire on it. And it may become a Death Star pretty soon.
ClosedAI being somewhat open for once, who would thought that ?
The models are way more crappier. Don't bother to use it