Sam Altman admits China’s DeepSeek forced OpenAI’s hand on open models: ‘If we didn’t do it…’

OpenAI released its first open-weight models since GPT-2 earlier this month, promising strong real-world performance at low cost. With China’s DeepSeek shaking up the AI industry earlier this year through open-source releases, it became clear why OpenAI didn’t want to be on what CEO Sam Altman called the “wrong side of history.” For the first time since the launch of its GPT-OSS models, Altman has openly acknowledged the “China factor” behind the move.

In an interaction with CNBC, Altman admitted that competition from Chinese open-source models like DeepSeek played a major role in OpenAI’s decision.

“It was clear that if we didn’t do it, the world was gonna head to be mostly built on Chinese open source models,” Altman told the publication.

“That was a factor in our decision, for sure. Wasn’t the only one, but that loomed large” he added

Altman also talked about the US policy of restricting the export of powerful semiconductors to China, stating, “My instinct is that doesn’t work,” he said.

“You can export-control one thing, but maybe not the right thing… maybe people build fabs or find other workarounds,” he added

DeepSeek was not alone, however, many other Chinese companies had been gaining prominence in the tech circles due to their open weights models. Alibaba’s Qwen for instance has been releasing its latest foundation models under the Apache 2.0 license. Meanwhile, Meta has already been shippng its Llama models uner community licenses while also adding these models directly on its social media platforms. 

Earlier in the year, the growing popularity of DeepSeek’s AI models shattered all notions of American supremacy in the AI race, as the chatbot showcased performance similar to rival models from OpenAI and Google, despite being developed at a fraction of the cost.

Meanwhile, OpenAI also released its latest GPT-5 model this month with claimed improvements in accuracy, reasoning, coding, health, writing and multi-modal abilities. The new model also led to the deprecation of older GPT and o series models for free users while the Pro, Plus, Team and Enterprise users still have access to the older models.