Qwen3 is Alibaba’s debut into so-called “hybrid reasoning models,” which it says combines traditional LLM capabilities with “advanced, dynamic reasoning.”
Sopa Images | Lightrocket | Getty Images
Alibaba released the next generation of its open-sourced large language models, Qwen3, on Tuesday — and experts are calling it yet another breakthrough in China’s booming open-source artificial intelligence space.
In a blog post, the Chinese tech giant said Qwen3 promises improvements in reasoning, instruction following, tool usage and multilingual tasks, rivaling other top-tier models such as DeepSeek’s R1 in several industry benchmarks.
The LLM series includes eight variations that span a range of architectures and sizes, offering developers flexibility when using Qwen to build AI applications for edge devices like mobile phones.
Qwen3 is also Alibaba’s debut into so-called “hybrid reasoning models,” which it says combines traditional LLM capabilities with “advanced, dynamic reasoning.”
According to Alibaba, such models can seamlessly transition between a “thinking mode” for complex tasks such as coding and a “non-thinking mode” for faster, general-purpose responses.
“Notably, the Qwen3-235B-A22B MoE model significantly lowers deployment costs compared to other state-of-the-art models, reinforcing Alibaba’s commitment to accessible, high-performance AI,” Alibaba said.
The new models are already freely available for individual users on platforms like Hugging Face and GitHub, as well as Alibaba Cloud’s web interface. Qwen3 is also being used to power Alibaba’s AI assistant, Quark.
China’s AI advancement
AI analysts told CNBC that the Qwen3 represents a serious challenge to Alibaba’s counterparts in China, as well as industry leaders in the U.S.
In a statement to CNBC, Wei Sun, principal analyst of artificial intelligence at Counterpoint Research, said the Qwen3 series is a “significant breakthrough—not just for its best-in-class performance” but also for several features that point to the “application potential of the models.”
Those features include Qwen3’s hybrid thinking mode, its multilingual support covering 119 languages and dialects and its open-source availability, Sun added.
Open-source software generally refers to software in which the source code is made freely available on the web for possible modification and redistribution. At the start of this year, DeepSeek’s open-sourced R1 model rocked the AI world and quickly became a catalyst for China’s AI space and open-source model adoption.
“Alibaba’s release of the Qwen 3 series further underscores the strong capabilities of Chinese labs to develop highly competitive, innovative, and open-source models, despite mounting pressure from tightened U.S. export controls,” said Ray Wang, a Washington-based analyst focusing on U.S.-China economic and technology competition.
According to Alibaba, Qwen has already become one of the world’s most widely adopted open-source AI model series, attracting over 300 million downloads worldwide and more than 100,000 derivative models on Hugging Face.
Wang said that this adoption could continue with Qwen3, adding that its performance claims may make it the best open-source model globally — though still behind the world’s most cutting-edge models like OpenAI’s o3 and o4-mini.
Chinese competitors like Baidu have also rushed to release new AI models after the emergence of DeepSeek, including making plans to shift toward a more open-source business model.
Meanwhile, Reuters reported in February that DeepSeek is accelerating the launch of its successor to its R1, citing anonymous sources.
“In the broader context of the U.S.-China AI race, the gap between American and Chinese labs has narrowed—likely to a few months, and some might argue, even to just weeks,” Wang said.
“With the latest release of Qwen 3 and the upcoming launch of DeepSeek’s R2, this gap is unlikely to widen—and may even continue to shrink.”
A vehicle Tesla is using for robotaxi testing purposes on Oltorf Street in Austin, Texas, US, on Sunday, June 22, 2025.
Tim Goessman | Bloomberg | Getty Images
In an earnings call this week, Tesla CEO Elon Musk teased an expansion of his company’s fledgling robotaxi service to the San Francisco Bay Area and other U.S. markets.
But California regulators are making clear that Tesla is not authorized to carry passengers on public roads in autonomous vehicles and would require a human driver in control at all times.
“Tesla is not allowed to test or transport the public (paid or unpaid) in an AV with or without a driver,” the California Public Utilities Commission told CNBC in an email on Friday. “Tesla is allowed to transport the public (paid or unpaid) in a non-AV, which, of course, would have a driver.”
In other words, Tesla’s service in the state will have to be more taxi than robot.
Tesla has what’s known in California as a charter-party carrier permit, which allows it to run a private car service with human drivers, similar to limousine companies or sightseeing services.
The commission said it received a notification from Tesla on Thursday that the company plans to “extend operations” under its permit to “offer service to friends and family of employees and to select members of the public,” across much of the Bay Area.
But under Tesla’s permit, that service can only be with non-AVs, the CPUC said.
The California Department of Motor Vehicles told CNBC that Tesla has had a “drivered testing permit” since 2014, allowing the company to operate AVs with a safety driver present, but not to collect fees. The safety drivers must be Tesla employees, contractors or designees of the manufacturer under that permit, the DMV said.
In Austin, Texas, Tesla is currently testing out a robotaxi service, using its Model Y SUVs equipped with the company’s latest automated driving software and hardware. The limited service operates during daylight hours and in good weather, on roads with a speed limit of 40 miles per hour.
Robotaxis in Austin are remotely supervised by Tesla employees, and include a human safety supervisor in the front passenger seat. The service is now limited to invited users, who agree to the terms of Tesla’s “early access program.”
On Friday, Business Insider, citing an internal Tesla memo, reported that Tesla told staff it planned to expand its robotaxi service to the San Francisco Bay Area this weekend. Tesla didn’t respond to a request for comment on that report.
In a separate matter in California, the DMV has accused Tesla of misleading consumers about the capabilities of its driver assistance systems, previously marketed under the names Autopilot and Full Self-Driving (or FSD).
Tesla now calls its premium driver assistance features, “FSD Supervised.” In owners manuals, Tesla says Autopilot and FSD Supervised are “hands on” systems, requiring a driver at the wheel, ready to steer or brake at all times.
But in user-generated videos shared by Tesla on X, the company shows customers using FSD hands-free while engaged in other tasks. The DMV is arguing that Tesla’s license to sell vehicles in California should be suspended, with arguments ongoing through Friday at the state’s Office of Administrative Hearings in Oakland.
Under California state law, autonomous taxi services are regulated at the state level. Some city and county officials said on Friday that they were out of the loop regarding a potential Tesla service in the state.
Stephanie Moulton-Peters, a member of the Marin County Board of Supervisors, said in a phone interview that she had not heard from Tesla about its plans. She urged the company to be more transparent.
“I certainly expect they will tell us and I think it’s a good business practice to do that,” she said.
Moulton-Peters said she was undecided on robotaxis generally and wasn’t sure how Marin County, located north of San Francisco, would react to Tesla’s service.
“The news of change coming always has mixed results in the community,” she said.
Brian Colbert, another member of the Marin County Board of Supervisors, said in an interview that he’s open to the idea of Tesla’s service being a good thing but that he was disappointed in the lack of communication.
“They should have done a better job about informing the community about the launch,” he said.
Alphabet’s Waymo, which is far ahead of Tesla in the robotaxi market, obtained a number of permits from the DMV and CPUC before starting its driverless ride-hailing service in the state.
Waymo was granted a CPUC driverless deployment permit in 2023, allowing it to charge for rides in the state. The company has been seeking amendments to both its DMV and CPUC driverless deployment permits as it expands its service territory in the state.
Meta CEO Mark Zuckerberg makes a keynote speech during the Meta Connect annual event, at the company’s headquarters in Menlo Park, California, on Sept. 25, 2024.
Manuel Orbegozo | Reuters
Meta CEO Mark Zuckerberg on Friday said Shengjia Zhao, the co-creator of OpenAI’s ChatGPT, will serve as the chief scientist of Meta Superintelligence Labs.
Zuckerberg has been on a multibillion-dollar artificial intelligence hiring blitz in recent weeks, highlighted by a $14 billion investment in Scale AI. In June, Zuckerberg announced a new organization called Meta Superintelligence Labs that’s made up of top AI researchers and engineers.
Zhao’s name was listed among other new hires in the June memo, but Zuckerberg said Friday that Zhao co-founded the lab and “has been our lead scientist from day one.” Zhao will work directly with Zuckerberg and Alexandr Wang, the former CEO of Scale AI who is acting as Meta’s chief AI officer.
“Shengjia has already pioneered several breakthroughs including a new scaling paradigm and distinguished himself as a leader in the field,” Zuckerberg wrote in a social media post. “I’m looking forward to working closely with him to advance his scientific vision.”
Read more CNBC tech news
In addition to co-creating ChatGPT, Zhao helped build OpenAI’s GPT-4, mini models, 4.1 and o3, and he previously led synthetic data at OpenAI, according to Zuckerberg’s June memo.
Meta Superintelligence Labs will be where employees work on foundation models such as the open-source Llama family of AI models, products and Fundamental Artificial Intelligence Research projects.
The social media company will invest “hundreds of billions of dollars” into AI compute infrastructure, Zuckerberg said earlier this month.
“The next few years are going to be very exciting!” Zuckerberg wrote Friday.
Alex Karp, CEO of Palantir Technologies, speaks on a panel titled Power, Purpose, and the New American Century at the Hill and Valley Forum at the U.S. Capitol on April 30, 2025 in Washington, DC.
Kevin Dietsch | Getty Images
Palantir has hit another major milestone in its meteoric stock rise. It’s now one of the 20 most valuable U.S. companies.
The provider of software and data analytics technology to defense agencies saw its stock rise more than 2% on Friday to another record, lifting the company’s market cap to $375 billion, which puts it ahead of Home Depot and Procter & Gamble. The company’s market value was already higher than Bank of America and Coca-Cola.
Palantir has more than doubled in value this year as investors ramp up bets on the company’s artificial intelligence business and closer ties to the U.S. government. Since its founding in 2003 by Peter Thiel, CEO Alex Karp and others, the company has steadily accrued a growing list of customers.
Revenue in Palantir’s U.S. government business increased 45% to $373 million in its most recent quarter, while total sales rose 39% to $884 million. The company next reports results on Aug. 4.
Buying the stock at these levels requires investors to pay hefty multiples. Palantir currently trades for 273 times forward earnings, according to FactSet. The only other company in the top 20 with a triple-digit ratio is Tesla at 175.
With $3.1 billion in total revenue over the past year, Palantir is a fraction the size of the next smallest company by sales among the top 20 by market cap. Mastercard, which is valued at $518 billion, is closest with sales over the past four quarters of roughly $29 billion.