OpenAI plans to once again live up to the adjective in its name.
The company announced that, “in the coming months,” it will release its first open-weights language model since GPT-2 in 2019.
The move comes as the splash caused by DeepSeek’s open models enlivened the open side of an ongoing split among AI developers. Meta’s release of its latest Llama 4 models this month, despite some controversy around benchmarking, has also amped up competition in the space.
Experts say OpenAI’s announcement speaks to growing competition from the open side of the debate, especially as businesses increasingly see open models as a gateway into generative AI or an alternative for highly regulated industries concerned about sensitive data.
OpenAI CEO Sam Altman hinted at such a release in the wake of DeepSeek’s R1 reasoning model in January.
“I personally think we have been on the wrong side of history here and need to figure out a different open-source strategy,” Altman wrote in a Reddit AMA that month.
Closing up: OpenAI’s pullback from its commitment to open-source began in 2019, when it declared its new GPT-2 model—the first that could generate convincingly realistic text from whole cloth—would be too dangerous in the wrong hands. It instead staggered the release throughout that year, while OpenAI converted from a nonprofit to a hybrid for-profit structure and soon took its first $1 billion investment from Microsoft.
With GPT-3’s debut in 2020, OpenAI launched its API and pitched the model as a commercial product for the first time.
The risks of open models in the hands of bad actors are still real, though, and OpenAI has reiterated that its API approach has allowed it to mitigate such dangers in the years since GPT-2.
An on-ramp: Yet open-weights models are also becoming an essential tool for business experimentation. Gartner VP Analyst Chirag Dekate told us most enterprises and developers start their generative AI exploration with open-weights models because they tend to be cheaper to run.
For AI companies, that means offering an open-weights model can help to seed their technology across developer tools and enterprise platforms, Dekate said.
“Dominating the commanding heights of the emerging GenAI economy requires you to embed your capabilities in diverse ecosystems, diverse agentic environments and experiences,” Dekate said. “Open-weights models from OpenAI enable and open up new opportunities for OpenAI to influence the evolving market ecosystem.”
Keep up with the innovative tech transforming business
Tech Brew keeps business leaders up-to-date on the latest innovations, automation advances, policy shifts, and more, so they can make informed decisions about tech.
Matthieu Lavergne, a partner at French VC firm Serena who just authored an investment report on open-source AI, said the move will help OpenAI attract developers and thus expand commercial opportunities.
“Developers want open-source products because they are superior, they are more composable. They have less vendor lock-in, there’s a community around it,” Lavergne told Tech Brew. “The more my models get spread to those developers, the larger my top of the funnel will be, and I will work commercially on those products.”
Erin Yepis, a senior analyst of market research and insights at Stack Overflow, told Tech Brew that she was surprised that open-weights models held their own with OpenAI’s models among developers in a recent survey.
“I expected OpenAI’s proprietary GPT-4o model to be popular because ChatGPT is widely used…but three of the top five LLMs surveyed were open source (sometimes referred to as open weight),” Yepis said in an email. “While many of the leading proprietary AI models now have the benefit of brand awareness and recognition, it seems developers continue to be interested in exploring and experimenting with new and alternative solutions as well.”
“Competitive pressure”: David Schubmehl, research VP of AI and automation at IDC, said the move is a sign of the competitive pressure OpenAI might be feeling with DeepSeek, Meta, Mistral, and Google’s Gemma all offering open-weights alternatives.
“I think the OpenAI folks were just feeling a little bit of pressure there that they needed to provide more capability for people to adjust and fine-tune and essentially customize the model, the OpenAI model, to fit that particular organization’s needs or that use case need,” Schubmehl told us.
Open-weights models are particularly useful for governments and highly regulated industries, like financial services and healthcare, where companies need customizable local solutions and can’t legally risk sensitive information in an API, experts say. Open-source models still lag in performance compared to proprietary ecosystems, however—Lavergne said they tend to be about one year behind.
But Dekate said he doesn’t expect this push and pull between the two camps to let up soon. “This tension between proprietary and open weights will remain. It will continue, and it’s a healthy balance.”