Open-source AI is making visible progress, but OpenAI CEO Sam Altman says that does not mean the company plans to open its GPT models. During a visit to TU Munich, he discussed the advances and the security questions that arise as models become more capable.
Open-source progress is changing the conversation
Altman described recent progress in open-source AI as “really good.” The source article points to work that can bring a chatbot close to ChatGPT using freely available models with little technical and financial effort. It also describes an internal paper by a Google engineer that considered how open-source models might overtake commercial offerings.
These developments mean that open-source systems are an increasingly serious part of the AI landscape. Altman appeared relaxed about that progress, while suggesting that its implications may depend on how capable models become. He said there may be a performance threshold beyond which open-source development needs to be considered differently.
OpenAI does not plan an open GPT-6 release
At TU Munich, Altman asked the audience whether they would like to see a potential GPT-6 released as open source immediately after training. Many people raised their hands. His answer was direct: “We're not going to do that, but it's interesting to hear that.”
That response concerns GPT models specifically. It does not rule out OpenAI releasing other projects as open source. The article names Whisper, the company’s speech-to-text model, as an example of an individual AI project that could still be released openly.
The distinction matters: Altman’s comments describe a position on highly capable GPT models, not a blanket end to every possible open-source release from OpenAI.
Why Altman links model access to security
Altman’s argument centers on the time between the introduction of an advanced model and the arrival of an open-source model with similar capabilities. He said that this lag could give developers time to improve security. In his view, the gap may become more significant as AI systems grow more powerful.
He also suggested that the delay will not be very long. Altman said, “The open-source lag won't be that much, and a few years later, when we know what the limits are and how to make it safe, there will be very good open-source models.” His comments frame the delay as a period for learning about limits and safety, rather than a permanent separation between commercial and open-source capabilities.
Altman had previously told the U.S. Senate that advanced models could have special capabilities, giving them a competitive advantage even in a strong open-source landscape. In Munich, he emphasized the security dimension of that difference: a lead in capability could create time to address risks before similar systems become broadly available.
A debate about timing and access
The comments point to a tension at the center of AI development. Open releases can make capable technology more accessible, and recent progress suggests that freely available models can narrow the distance to commercial systems. At the same time, Altman argues that controlling the timing of releases for the most advanced GPT models could help with security work.
His remarks leave open questions about where a performance threshold would sit and how long the gap between commercial and open-source models might last. The article does not specify a threshold or a release schedule. What it does establish is Altman’s current position: OpenAI will not open-source a potential GPT-6 immediately after training, while the future of other projects such as Whisper remains distinct from that decision.