Surveying the state of America’s artificial intelligence landscape earlier this year, Misha Laskin was concerned.
Laskin, a theoretical physicist and machine learning engineer who helped create some of Google’s most powerful AI models, saw a growing embrace among American AI companies of free, customizable and increasingly powerful “open” AI models.
But most of these models were being made in China, and these systems were quickly gaining ground on their U.S. competitors.
“These models were not that far behind the frontier. In fact, they were surprisingly close to the frontier. The ones that are coming now,” Laskin said, pausing slightly, “well they’re palpably close to the frontier.”
Laskin founded a startup called Reflection AI, recently valued at $8 billion, to provide an open-source American alternative to these increasingly capable Chinese models that have gained traction in Silicon Valley.
“You’re starting to see glimpses of open-model companies actually driving the frontier of intelligence in China, and overall, the frontier of intelligence,” Laskin said.
Over the past year, a growing share of America’s hottest AI startups have turned to open Chinese AI models that increasingly rival, and sometimes replace, expensive U.S. systems as the foundation for American AI products.
NBC News spoke to over 15 AI startup founders, machine-learning engineers, industry experts and investors, who said that while models from American companies continue to set the pace of progress at the frontier of AI capabilities, many Chinese systems are cheaper to access, more customizable and have become sufficiently capable for many uses over the past year.
The growing embrace could pose a problem for the U.S. AI industry. Investors have staked tens of billions on OpenAI and Anthropic, wagering that leading American artificial intelligence companies will dominate the world’s AI market. But the increasing use of free Chinese models by American companies raises questions about how exceptional those models actually are — and whether America’s pursuit of closed models might be misguided altogether.
Michael Fine, head machine learning at Exa, an AI-focused search company valued at $700 million and supported by Silicon Valley mainstays like Lightspeed Venture Partners and Nvidia, said running Chinese models on Exa’s own hardware has proved to be significantly faster and less expensive than using bigger models, like OpenAI’s GPT-5 or Google’s Gemini, in many cases.
“What often happens is we’ll get a feature working with a closed model and realize it’s too expensive or too slow, and we ask, ‘What levers do we have to make this faster and cheaper?’”
“That usually means replacing the closed model with the equivalent open model and then running it on our own infrastructure,” Fine said.
Chinese models, like DeepSeek’s R1 and Alibaba’s Qwen, are free to use and considered “open-source” or “open-weight” because anyone can download, copy, modify and operate them. They differ from leading American systems like Anthropic’s Claude or OpenAI’s most popular GPT models, which are “closed,” or proprietary, and accessed through data centers and pipelines controlled by the big tech giants.
For years, American closed-source models from OpenAI and Anthropic vastly outperformed both American and Chinese open alternatives. Even well-resourced in-house efforts to use open-source models struggled: Bloomberg tried to create an internal tool, BloombergGPT, using open-source models trained on its expansive collection of financial news and documents, only to see it trail OpenAI’s closed models on financial knowledge.
Yet in the past year, Chinese companies like DeepSeek and Alibaba have made huge technological advancements. Their open-source products now closely approach or even match the performance of leading closed American models in many domains, according to metrics tracked by Artificial Analysis, an independent AI benchmarking company.
“The gap is really shrinking,” Lin Qiao, CEO of Fireworks AI and co-creator of PyTorch, the dominant framework for training AI models, said of the capability differences between American closed-source and Chinese open-source models.
https://iframe.nbcnews.com/mn2u6CU4?_showcaption=true&app=1
As a result of this increase in performance, some platforms that allow users to choose between different models, like OpenRouter, are seeing people gravitate toward Chinese open-source models.
Jerry Liu, the founder of Dayflow — a productivity app — estimates that roughly 40% of Dayflow’s users now choose to use open-source models.
Dayflow is built around a few core tasks like scanning screenshots and summarizing users’ activity. The app lets users choose from several AI models to complete these tasks, including Google’s Gemini and smaller open-source options such as Alibaba’s Qwen.
For tasks like describing a user’s screen, Liu said that Qwen is remarkably consistent. “Qwen is as good as GPT-5 for my use case,” Liu said.
And unlike GPT-5 or Gemini, a smaller version of Qwen can run at a relatively low cost or for free. Liu said paying for users’ closed-model usage can cost Dayflow up to $1,000 per person, making the cheaper open-source models important to Dayflow’s viability.
Dayflow’s open-source models also perform all of their processing on each user’s individual computer, which Liu said is appealing to those who don’t want to send their data to the cloud for privacy reasons. Liu’s personal preference is to keep things on his device using open-source models: “Would I use a product where my entire screen was beaming up to some random guy’s cloud? Hell no.”
Beyond increased performance, stronger privacy and lower cost, open-source models are also gaining ground through ecosystem advantages. Increasing open-source adoption and the creation of open-source resources by developers are encouraging more developers to use these models.
Antonio Vespoli, co-founder of browser agent startup Circlemind AI, said Chinese models now dominate developer resources online. The reason is practical: Chinese models like Qwen — which Airbnb “heavily” relies on, according to CEO Brian Chesky — have abundant training guides and community support.
Charles Zedlewski, chief product officer at AI infrastructure company Together AI, said developers now find it simpler and more efficient to start from open models and adapt them with their own data, adding “skills or knowledge that aren’t readily available in any of the models out there today.” As companies ship their first AI applications, he said, they’re getting a clear sense of their needs.
For developers looking to customize models, those resources make Chinese options the default starting point. Kilo Code, a popular coding app that helps developers write software with AI, lets users choose from a variety of models. Of the 20 top models among Kilo Code users, seven are Chinese models, with six of those seven being open-source.
https://iframe.nbcnews.com/XIpsSHQ5?_showcaption=true&app=1
China embraces open-source
Whereas much of America’s AI development has taken place in the private sector, led by industry titans like OpenAI and Anthropic and their pursuit of a closed-model approach, China’s government has been more actively engaged in charting out the country’s AI vision.
In a Nov. 1 economic address, Chinese President Xi Jinping called for greater “cooperation on open-source technologies.” And in March, China’s top economic planning authority signaled its intent to support an ecosystem of open-source models.
Chinese labs generally release their models openly, while American companies like OpenAI saw early success with closed models and have stuck with the closed-source approach.
https://iframe.nbcnews.com/Txdn9nJk?_showcaption=true&app=1
Many Chinese companies also introduce products at a faster pace than their American counterparts: Alibaba released a new model roughly every 20 days this year, compared with Anthropic’s 47-day average between releases.
Nathan Lambert, a senior research scientist at the Allen Institute for AI and an expert on the open-model AI ecosystem, told NBC News that Chinese models’ recent progress is no fluke.
“The Chinese are genuine innovators in AI,” Lambert said.
The “balance of power has been shifting rapidly in the last 12 months,” Lambert added. He has written extensively about China’s AI developments on his Substack and is considered an expert on China’s open-source ecosystem.
America’s AI edge
Some in Silicon Valley are quick to point out that American models retain a significant advantage at the cutting edge of AI capabilities, and that these closed American models provide off-the-shelf convenience and ease of use that unwieldy open models cannot match.
Tim Tully, a partner at Silicon Valley venture capital firm Menlo Ventures, said closed models remain significantly more capable and often more useful: “The tooling is just better, the productivity is just better, the agentic frameworks that are built and used by everybody, they’re just better off with Anthropic and OpenAI. They just work better. So the ecosystem is just strong in the closed-sourced environment.”
In addition, many companies may shy away from using Chinese models because of risk — whether actual or imagined — about using a product based on a Chinese foundation.
“There’s a perceived risk that buyers are hesitant to buy a product that’s based on a Chinese open-weight model, either from the private sector or the public sector,” Tully said. Menlo Ventures is an investor in Anthropic, one of the world’s leading closed-model companies.
In late September, the U.S. Center for AI Standards and Innovation released a report outlining risks from DeepSeek’s popular models, finding weakened safety protocols and increased pro-Chinese outputs compared to American closed-source models.
A recent White House memo also accused Alibaba, Qwen’s developer, of supporting China’s military, adding a political barrier to enterprise embrace of these AI systems.
In reply, Alibaba told the Financial Times the assertions were “complete nonsense” and “plainly an attempt to manipulate public opinion and malign Alibaba.”
Many observers also note that several Chinese models released over the past year appear to have borrowed heavily from American models. Some observers think DeepSeek’s rapid progress could only have come from copying much of the difficult foundational work of American companies like OpenAI and Anthropic.
This dynamic raises questions about whether Chinese open models will continue to converge on, let alone eclipse, American closed-model performance. Over the past year, experts have charged that Chinese models may remain highly capable “fast-followers,” reliant on American AI progress.
Chinese firms, meanwhile, are also exploring closed-source models. In October, Alibaba released only a closed-source version of the largest of its new Qwen systems, opting not to share an open-source version.
Who controls the future?
American AI companies and the federal government have noticed the recent rise of Chinese models, and experts have even labeled America’s lack of powerful open-source models an “existential” threat to democracy.
While Meta’s high-profile Llama series of open-source models has historically led American open-source efforts, CEO Mark Zuckerberg has signaled Meta’s intention not to open-source all of its “superintelligence” AI models. The performance of Llama models has also stalled in recent years, one of the reasons why open-source users have shifted to better-performing Chinese open-source models.
Yet American open-source efforts may be gradually awakening, as American innovators attempt to boost American open-model competitiveness.
In July, the White House released an AI Action Plan that called for the federal government to “Encourage Open-Source and Open-Weight AI.”
In August, ChatGPT maker OpenAI released its first open-source model in five years. Announcing the model’s release, OpenAI cited the importance of American open-source models, writing that “broad access to these capable open-weights models created in the US helps expand democratic AI.”
And in late November, the Seattle-based Allen Institute released its newest open-source model called Olmo 3, designed to help users “build trustworthy features quickly, whether for research, education, or applications,” according to its launch announcement.
Lambert, of the Allen Institute, has also launched the “ATOM Project” — an acronym for “American Truly Open Models.” As the ATOM Project’s manifesto declares: “America has lost its lead in open models — both in performance and adoption — and is on pace to fall further behind.”
“If we want to be the preeminent nation in the AI era, we cannot cede such a critical piece of the ecosystem to any nation,” Lambert told NBC News via email.
Source: https://www.nbcnews.com/
