OpenAI's new models reshape the race with China

State of play: OpenAI's new models are designed for customers who want the cost savings and privacy that come from running AI models directly on their own devices rather than relying on cloud-based services like ChatGPT or its rivals.

• The company is also pitching the models to countries seeking greater control, local data storage and independence from cloud providers like Google and Microsoft.

What they're saying: OpenAI CEO Sam Altman stressed the political importance of the release, and industry leaders were quick to echo his sentiments.

• "We are excited for the world to be building on an open AI stack created in the United States, based on democratic values, available for free to all and for wide benefit," Altman said in a statement.
• Box CEO Aaron Levie warned that if U.S. companies fall behind, Chinese firms could dominate with open-source models optimized for Huawei chips. "It's important that America stays in the game," Levie said. "And it's great that OpenAI is taking the lead on that."
• Hugging Face CEO Clement Delangue called the release "critically important," pointing to Trump's AI Action Plan's call for stronger American open source AI foundations.
• Developer Simon Willison called the models "very impressive" and "eyebrow-raising," noting he hadn't expected open-weight models of this size to perform so well.

For Amazon and cloud providers beyond Microsoft, the move lets them offer access to OpenAI models for the first time.

• "It does look like a very impressive model that competes very, very well with everything out there," AWS VP David Brown told Axios. "It's very similar to the o4-mini model, which is a very capable model."
• While Amazon will charge customers for the computing cost of serving up the new models (as it does with Llama and other open source models), Brown said Amazon expects the new OpenAI models to offer twice as much performance for the price as OpenAI's comparable o4 models, three times as much performance for the price of a comparable Google Gemini model and five times that of DeepSeek.
• "This is definitely going to be something that I think customers are going to have a very keen interest in," he said.

Driving the news: Both of OpenAI's new models are capable of chain-of-thought reasoning and accessing the web.

• The first, a 117 billion parameter model called gpt-oss-120b, can run on a single GPU with 80 gigabytes of RAM.

• The second, with 21 billion parameters and called gpt-oss-20b, is designed to run on laptops or other devices with 16 gigabytes of RAM.
• Both models are available via the open source hosting platform Hugging Face and cloud providers, including Amazon. Microsoft is also offering a version of the smaller model that has been optimized to run on Windows devices.
• The company provided various benchmarks showing the open models performing at or near the performance of the company's o3 and o4-mini models.

Yes, but: The new open models are text-only, as compared with most of OpenAI's recent models, which are capable of processing and outputting text, images, audio and video.

• OpenAI wouldn't commit to a specific schedule for future open models. OpenAI hasn't released an open large language model since GPT-2 in 2019.

Between the lines: Technically, the models are "open weights" versus "open source," meaning anyone can download and fine-tune the models, but there's no public access to other key information, like training data details.

• That's similar to DeepSeek and many of Meta's Llama models, but not as open as OLMo from the Allen Institute for AI.
• OpenAI declined to comment to Axios on what the new models were trained on or how the training may differ from that of its closed models.

What to watch: Although the new models are competitive with current OpenAI models, the company's newest model — GPT-5 — is rumored to arrive in the coming days.