OpenAI's release of two open models yesterday propels the U.S. forward in its AI race with China, industry leaders told Axios.

The big picture: The arrival of China's DeepSeek model earlier this year — combined with Meta's refocusing of its open source efforts — had intensified concerns that China's open models could end up dominating the global market.

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."
• Brown noted that Amazon expects the new OpenAI models to cost its customers half or less what they pay to use competing models.

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.

Anthropic plans to offer its products to the federal government for as little as $1, per a source close to the AI company.

Why it matters: America's leading AI companies are now competing for the massive client that is the U.S. government, and how the government uses AI can set the tone for other industries and businesses.

• "We will continue to work with our partners across the government to expand Claude access for federal teams by streamlining procurement processes and offering unique pricing," a blog post from Anthropic reads.

Driving the news: Anthropic's Claude, Google's Gemini and OpenAI's ChatGPT were added to the U.S. government's Multiple Award Schedule yesterday, meaning they will become available for use at federal agencies.

The source close to Anthropic said the pricing could vary by customer or agency, and that the government hasn't yet been informed of the offer.

• Anthropic intends to offer what would otherwise be multiple paid accounts to the federal government for a total of $1, the source said.

If you follow markets and the economy, it's the only thing you've heard over the last few days: AI spending is propping everything up, from GDP to the S&P, and that can't last forever.

The big picture: Like every land rush in history, from railroads to websites with sock puppet mascots, investors' faith in AI is strong enough to ignore all the warning signs around them.

• Stripping out the data center boom, underlying economic growth isn't that strong.
• Valuations in the stock market are stretched beyond all historic norms, and the best earnings growth is coming from just a few megacap companies.

What they're saying: "While you can quibble with this or that estimate or figure, the order of magnitudes make it pretty clear that AI itself is not enough to buoy the economy, sales, or earnings ahead," Unlimited Funds chief investment officer Bob Elliott wrote in a Substack post Tuesday.

• "Instead it is going to require an overall economic productivity rise to meet the bulls hopes (and then some)."

The intrigue: AI has, so far, been very good at getting companies to spend on things. Buildings, chips, electricity, etc. The spend on all that stuff is running into the hundreds of billions of dollars a year.

• But it's not translating to spending on people. As Ritholtz Wealth Management's Callie Cox put it this week, "The human economy is stumbling, but the robot economy is flourishing."
• In fact, some employers are already touting how many humans they've been able to fire with AI. Less so how many they've been able to keep paying and retrain into more productive work.

Reality check: There's no sign yet that any of the biggest players in the market are at all inclined to pull back on their spending.

• While much of the attention was on OpenAI, Google dropped Genie 3, a model that creates virtual worlds from text prompts in real time. (The Verge)
• Meanwhile, Anthropic released Claude Opus 4.1, its latest code-generating ace. (VentureBeat)

I'm not sure I want or need these Windows XP Crocs, but apparently they are a real thing, as is this new collaboration between Timberland and Sonic the Hedgehog.