Are you looking for smarter insights delivered straight to your inbox? Sign up for our weekly newsletters to receive essential updates tailored for enterprise AI, data, and security leaders. Subscribe now!
OpenAI’s Return to Open Source
OpenAI has announced a significant shift back to its roots as an open-source AI company with the release of two new frontier large language models (LLMs): gpt-oss-120b and gpt-oss-20b. The gpt-oss-120b model features 120 billion parameters and can operate on a single Nvidia H100 graphics processing unit (GPU). In contrast, the gpt-oss-20b model, with 20 billion parameters, is compact enough to run on consumer laptops or desktop PCs.
Model Capabilities
Both models are text-only, meaning users can only input text messages and receive text responses. While they do not support file or image uploads like the multimodal AI systems available for the past two years, these models can still write code and solve mathematical problems. Notably, their performance on various tasks surpasses that of some of OpenAI’s paid models and many competitors worldwide.
Upcoming AI Impact Series Event
The next phase of AI is upon us, and you won’t want to miss it. Join industry leaders from Block, GSK, and SAP for an exclusive discussion on how autonomous agents are transforming enterprise workflows—from real-time decision-making to end-to-end automation. Secure your spot today, as space is limited: https://bit.ly/3GuuPLF.
Integration with External Tools
These models can also connect to external tools, including web search capabilities, allowing them to conduct research on behalf of users.
Free and Accessible
One of the most compelling aspects of these models is that they are free and available for both enterprises and independent developers. Users can download the code and modify it to suit their needs, running it locally without requiring an internet connection. This ensures maximum privacy, especially compared to other leading models from OpenAI and competitors like Google and Anthropic. The models can be accessed today with full weights through the AI code-sharing communities Hugging Face and GitHub.
Performance Benchmarks
According to OpenAI, the gpt-oss-120b model matches or exceeds the performance of its proprietary o4-mini model in various reasoning and tool-use benchmarks, including competition mathematics (AIME 2024 & 2025), general problem solving (MMLU and HLE), agentic evaluations (TauBench), and health-specific assessments (HealthBench). The smaller gpt-oss-20b model is comparable to o3-mini and outperforms it in several benchmarks.
Multilingual Capabilities
These models are multilingual and demonstrate strong performance across various non-English languages, although OpenAI has not disclosed specific details regarding which languages are supported. While the models are ready to use out of the box, OpenAI emphasizes that localized fine-tuning—such as its collaboration with the Swedish government to create a language-specific version—can significantly enhance performance in specific regional or linguistic contexts.
Licensing Terms
The licensing terms for both models are particularly noteworthy. They are released under the Apache 2.0 license, which is more enterprise-friendly than Meta’s complex Llama license, which requires a paid license for services with over 700 million monthly active users. In contrast, OpenAI’s gpt-oss series imposes no such restrictions. This allows consumers, developers, and enterprises of all sizes to download, fine-tune, and modify the gpt-oss models freely, enabling them to generate revenue or provide paid services without incurring costs to OpenAI.
Privacy and Security Considerations
Importantly, enterprises can utilize these powerful OpenAI models on their own hardware, ensuring complete privacy and security without transmitting any data to the cloud or external servers. This feature is especially crucial for highly regulated industries such as finance, healthcare, and legal services, as well as organizations in military, intelligence, and government sectors.
Before this announcement, users of ChatGPT or its application programming interface (API) were sending data to OpenAI servers, which could potentially be accessed by government agencies without user knowledge. This concern remains for anyone utilizing ChatGPT or the API moving forward, as OpenAI co-founder Sam Altman recently cautioned.