Artificial Intelligence pioneer OpenAI reveals the unveiling of two novel open-source AI models
In a significant move towards open AI development, OpenAI, the multibillion-dollar company, has released two new free and open-source AI models: gpt-oss-120b and gpt-oss-20b. This marks OpenAI's first fully open-weight model release since GPT-2, which was unveiled in 2019-2020.
Unlike its previous closed-weights releases and the upcoming GPT-5, these new models come with publicly accessible weights under the Apache 2.0 license, allowing commercial use without payment or permission. However, OpenAI is not releasing the training data or full architectural details for these models.
Key Features of gpt-oss-120b and gpt-oss-20b
These new models focus on text reasoning and chain-of-thought processes, trained using high-compute reinforcement learning (RL). They are text-only and do not handle images or audio modalities, differentiating them from multimodal models like DALL-E.
The models were trained on NVIDIA H100 GPUs and optimized for inference on NVIDIA's Blackwell platform, achieving high efficiency (1.5 million tokens/sec), aiming to democratize access to state-of-the-art AI capabilities across industries and scales.
Comparison with Prior and Future Models
| Feature | GPT-2 (2020) | gpt-oss-120b / gpt-oss-20b (2025) | GPT-5 (expected 2025) | |----------------------------------|---------------------------------------|---------------------------------------------------------|-----------------------------------------------| | Release Type | Open weights and architecture | Open weights; proprietary architecture & training data | Closed & proprietary | | Model Size | Smaller (~1.5B parameters) | Large (20B and 120B parameters) | Expected larger and more advanced | | Multimodal? | Text only | Text only | Likely multimodal | | Licensing | Open source | Apache 2.0 (free for commercial use) | Closed source | | Training Data Released? | Not fully public | No | Not public | | Primary Focus | Language generation | Reasoning with reinforcement learning | More general AI capabilities, likely stronger reasoning and multimodal | | Use Cases | Research & experimentation | Broad commercial, research, AI agents | Advanced AI applications, commercial products |
OpenAI's Renewed Commitment to Open AI Development
OpenAI's motivation for releasing these open-weight models links to democratizing AI access and encouraging innovation built on "American rails" with democratic values, contrasting with some competing international efforts. The open models also provide options for organizations with stringent data-residency or security requirements, avoiding cloud-based third-party AI.
OpenAI's CEO, Sam Altman, stated that he hopes the AI will assist with "new kinds of research and the creation of new kinds of products." The company has provided a feedback portal and a more extensive blog to explain the new models and how they work.
In summary, gpt-oss-120b and gpt-oss-20b are significant as OpenAI’s renewed commitment to open AI development, presenting large, powerful AI models with available weights but withholding sensitive training details, balancing openness with competitive and legal considerations. They differ from GPT-2 by scale and commercial licensing and from GPT-5 by openness and feature scope.
Elon Musk, known for his interest in artificial intelligence, may find the recent announcement by OpenAI intriguing. With these new models, gpt-oss-120b and gpt-oss-20b, individuals and businesses can utilize advanced artificial-intelligence technology easily while maintaining privacy, as these models come with publicly accessible weights under the Apache 2.0 license, allowing for commercial use without payment or permission, but not releasing the training data or full architectural details.
