Having initially stated that OpenAI was potentially “on the wrong side of history” with respect to open sourcing its technology, CEO Sam Altman announced the AI firm will release its first “open weight” language model since GPT-2 later this year. This comes as the company battles perceptions that it might be ceding territory to rivals such as DeepSeek and Meta, who have launched some versions of open weight AI models.“We’ve been thinking about this for a long time but other priorities took precedence. now it feels important to do,” Altman said in a post on X. The model will have “reasoning” capabilities along the lines of OpenAI’s o3-mini.
Open weight vs open source AI model
To be certain, Altman explained that OpenAI would indeed release an open weight model, and not an open source model — the former is less transparent than the latter.
Open source models are completely transparent, publishing source code, model structure, training algorithms, and weights under a licence permitting free usage, modification, and distribution. Preferably, training data is made available, but legal restrictions usually prevent this. Open weight models only publish the trained model weights, but not the source code, training data, or complete architecture information. This limits transparency and customisation, as users can execute the model but not completely transform or retrain it.
Why OpenAI is changing direction
Following years of concentrating on closed source technology, the change in direction at OpenAI was necessitated by the launch of China’s DeepSeek. The latter demonstrated to the world that it was possible to create a language model, an open sourced model, at a fraction of what it cost some of its rivals to create a model. Meta also found success in its open-weight model, Llama, that achieved over one billion downloads — although developers were vociferous about its model’s licence conditions possibly being too restrictive for commerce.
OpenAI currently offers its AI models through a chatbot and cloud, as opposed to its competitors whose models people are able to download and re-work.
At an open Reddit Q&A session earlier, OpenAI CEO Sam Altman explained that the company had previously been in the wrong in regards to open sourcing its technology. “[Personally I believe that we need to] come up with a different open source plan,” Altman replied. Not everyone within OpenAI feels this way, and it’s not our highest priority these days either… We will create superior models, but we will have less of an advantage than we had in past years.
Based on a feedback form that OpenAI posted on its website, the firm was soliciting “developers, researchers, and members of the broader community” and asked questions such as, “What would you like to see in an open weight model from OpenAI?” and “What open models have you used in the past?”
When will OpenAI open-source its open weight AI model?
In his X post, Altman explained that prior to launch, OpenAI would test this model based on its preparedness framework, “similar to any other model. And we will do additional work since we already know that this model will be adapted after release.”.
The company still has “some decisions to make” and is hosting developer events to gather feedback and later play with early prototypes. “We’ll start in (San Francisco) in a couple of weeks followed by sessions in Europe and APAC,” he added.
