OpenAI to launch its first 'open-weights' model since 2019

OpenAI is planning to release its first "open-weights" model since 2019, aiming to adopt a more open strategy in response to competitive pressures. Unlike fully open-source models, open-weights models make their weights transparent and publicly available, allowing developers to customize them without retraining on new data. This approach offers a middle ground between open- and closed-source models, enabling easier and cheaper customization for specific tasks. OpenAI's move comes as it faces competition from companies like DeepSeek and Meta, which are also developing models with varying degrees of openness.

The new model will feature reasoning capabilities similar to OpenAI's o3-mini model and is expected to compete with Meta's Llama family and DeepSeek's V3 model. OpenAI has created a feedback form for developers to provide input on the new model's features. While the open-weights model will not disclose the underlying code or training data, it represents a shift towards more transparency in AI development. OpenAI CEO Sam Altman mentioned that the model will undergo evaluation according to the company's "preparedness framework" before its release, with no specific launch date announced yet.

Key takeaways:

OpenAI plans to release its first "open-weights" model since 2019, allowing developers to see and alter the model's weights for customization.
The open-weights model will have "reasoning" capabilities similar to OpenAI's o3-mini model, enhancing response accuracy.
Open-weights models offer a middle ground between open- and closed-source models, providing transparency in weights but not in source code or training data.
OpenAI's move towards open-weights models is a strategic response to competition from rivals like Meta's Llama and DeepSeek's V3 model.

OpenAI to launch its first 'open-weights' model since 2019 - SiliconANGLE

Key takeaways:

Comments (0)

Newsletter