Alibaba unveils Qwen 3, a family of 'hybrid' AI reasoning models
Chinese tech company Alibaba on Monday released Qwen 3, a family of AI models the company claims matches and in some cases outperforms the best models available from Google and OpenAI.
Most of the models are — or soon will be — available for download under an "open" license from AI dev platform Hugging Face and GitHub. They range in size from 0.6 billion parameters to 235 billion parameters. Parameters roughly correspond to a model's problem-solving skills, and models with more parameters generally perform better than those with fewer parameters.
The rise of China-originated model series like Qwen have increased the pressure on American labs such as OpenAI to deliver more capable AI technologies. They've also led policymakers to implement restrictions aimed at limiting the ability of Chinese AI companies to obtain the chips necessary to train models.
According to Alibaba, Qwen 3 models are "hybrid" models in the sense that they can take time and "reason" through complex problems or answer simpler requests quickly. Reasoning enables the models to effectively fact-check themselves, similar to models like OpenAI's o3, but at the cost of higher latency.
"We have seamlessly integrated thinking and non-thinking modes, offering users the flexibility to control the thinking budget," wrote the Qwen team in a blog post.
The Qwen 3 models support 119 languages, Alibaba says, and were trained on a data set of nearly 36 trillion tokens. Tokens are the raw bits of data that the model processes; 1 million tokens is equivalent to about 750,000 words. Alibaba says Qwen 3 was trained on a combination of textbooks, "question-answer pairs," code snippets, and more.
These improvements, along with others, greatly boosted Qwen 3's performance compared to its predecessor, Qwen 2, says Alibaba. On Codeforces, a platform for programming contests, the largest Qwen 3 model — Qwen-3-235B-A22B — beats out OpenAI's o3-mini. Qwen-3-235B-A22B also bests o3-mini on the latest version of AIME, a challenging math benchmark, and BFCL, a test for assessing a model's ability to "reason" about problems.
But Qwen-3-235B-A22B isn't publicly available — at least not yet.
The largest public Qwen 3 model, Qwen3-32B, is still competitive with a number of proprietary and open AI models, including Chinese AI lab DeepSeek's R1. Qwen3-32B surpasses OpenAI's o1 model on several tests, including an accuracy benchmark called LiveBench.
Alibaba says Qwen 3 "excels" in tool-calling capabilities as well as following instructions and copying specific data formats. In addition to releasing models for download, Qwen 3 is available from cloud providers including Fireworks AI and Hyperbolic.
Tuhin Srivastava, co-founder and CEO of AI cloud host Baseten, said that Qwen 3 is another point in the trend line of open models keeping pace with closed-source systems such as OpenAI's.
"The U.S. is doubling down on restricting sales of chips to China and purchases from China, but models like Qwen 3 that are state-of-the-art and open [...] will undoubtedly be used domestically," he told TechCrunch in a statement. "It reflects the reality that businesses are both building their own tools [as well as] buying off the shelf via closed-model companies like Anthropic and OpenAI."
This article originally appeared on TechCrunch at https://techcrunch.com/2025/04/28/alibaba-unveils-qwen-3-a-family-of-hybrid-ai-reasoning-models/
Hashtags
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Yahoo
35 minutes ago
- Yahoo
Bill to ban TikTok from state-owned devices in Pennsylvania advances
(WHTM) — Multiple cybersecurity bills, including one banning popular social media platform TikTok from state-owned devices in Pennsylvania, have advanced out of the Senate Communications and Technology Committee. Senate Bill 376, sponsored by state Senator Kristin Phillips-Hill (R-York), would ban TikTok on all state-owned devices and networks. According to the memo, this comes as a response to growing national security concerns. Close Thanks for signing up! Watch for us in your inbox. Subscribe Now The bill would require Pennsylvania to follow federal guidelines regarding the use of TikTok on Government devices and networks. Many politicians across the United States consider TikTok a national security risk because the platform is owned by the Chinese internet technology company ByteDance. Senator Phillips-Hill's second proposed bill, Senate Bill 377, would require all state IT contracts to meet the cybersecurity standards set by the National Institute of Standards and Technology. According to the memo, this would reduce the risk of data breaches across state systems. 'Back in March, I made it clear that cybersecurity would be a top priority this session, and today, we're taking meaningful steps forward,' Sen. Phillips-Hill said. 'These measures reflect best practices that other states and the federal government are already implementing to protect against bad actors. If we want to stay ahead of cyber threats, we need to take proactive, proven steps to secure our data and defend our systems.' The bills are part of a larger legislative package introduced by Sen. Phillips-Hill that intends to better safeguard Pennsylvania's information technology. The bill will now go to the Senate for consideration. Copyright 2025 Nexstar Media, Inc. All rights reserved. This material may not be published, broadcast, rewritten, or redistributed.
Yahoo
35 minutes ago
- Yahoo
TikTok ban on Pa. state government devices passes Senate
HARRISBURG, Pa (WHTM) — A bill banning TikTok on state-owned devices in Pennsylvania unanimously passed the Senate Wednesday. Senate Bill 376, sponsored by state Senator Kristin Phillips-Hill (R-York), passed with unanimous support. It would bring the Commonwealth in line with more than 39 states that passed similar restrictions in the wake of security concerns over the app's parent company, ByteDance, and its relationship with the Chinese government. Close Thanks for signing up! Watch for us in your inbox. Subscribe Now 'If we want to protect Pennsylvanians from cyber attacks, we must take proactive and proven steps to defend our systems against known vulnerabilities and threats,' said Sen. Phillips-Hill. 'These bills reflect commonsense safeguards that are already being used by other states and the federal government.' The bill would require state employees to follow federal guidelines on the use of TikTok on government networks and devices. Two years ago, State Treasurer Stacy Garrity banned TikTok on Pennsylvania Treasury devices. What is TikTok shop, and where does all the stuff come from? The Senate also unanimously passed Senate Bill 377, which would require all state IT contracts to meet cybersecurity standards established by the National Institute of Standards and Technology. Both bills are a part of a larger package from Sen. Phillips-Hill and now advance to the House for consideration. Copyright 2025 Nexstar Media, Inc. All rights reserved. This material may not be published, broadcast, rewritten, or redistributed.
Yahoo
35 minutes ago
- Yahoo
Former Google CEO Eric Schmidt's AI Expo serves up visions of war, robotics, and LLMs for throngs of tech execs, defense officials, and fresh recruits
Drones buzz overhead, piercing the human hum in the crowded Walter E. Washington Convention Center. On the ground, tech executives, uniformed Army officers, policy wonks, and politicians compete for attention as swarms of people move throughout the vast space. There are pitches about 'next generation of warfighters,' and panels about winning the 'AI innovation race.' There are job seekers and dignitaries. And at the center of it all, there is Eric Schmidt. The former Google CEO is the cofounder of the non-profit that organizes the confab, known as the AI+ Expo for National Competitiveness. The Washington DC event, now in its second year, is a fascinating, real world manifestation of the Schmidt worldview, which sees artificial intelligence, business, geopolitics, and national defense as interconnected forces reshaping America's global strategy. If it was once the province of futurists and think tank researchers, this worldview is now increasingly the consensus view. From innovations like self-driving robotaxis circulating in multiple U.S. cities to Silicon Valley defense startups snagging large government contracts, evidence of change—and of the stakes involved—are everywhere. The AI+ Expo, hosted by Schmidt's Special Competitive Studies Project (SCSP), is ground zero for the stakeholders of this new world order: Thousands of Washington insiders, military brass, tech executives, policymakers, students, and curious professionals are drawn together under one stars and stripes-style banner—to ensure America leads the AI age. It's a kind of AI county fair, albeit one where Lockheed Martin stages demos, defense tech companies hand out swag, Condoleezza Rice takes the stage, and protesters outside chant 'No tech for genocide' during Schmidt's keynote. The event is free to attend, and lines to enter stretch around the block. The only thing missing? GPU corn dogs on a stick, perhaps. Soft lobbying is omnipresent at the event. Tesla, for instance, offers self-driving tech demos just as the company prepares to launch its first robotaxi service in Austin and as Elon Musk pushes lawmakers on autonomous vehicle regulations. OpenAI is also working the room, touting its o3 reasoning model, newly deployed on a secure government supercomputer at Los Alamos. 'The transfer of model weights occurred via highly secured and heavily encrypted hard drives that were hand carried by OpenAI personnel to the Lab,' a company spokesperson told Fortune. 'For us, today's milestone, and this partnership more broadly, represents more than a technical achievement. It signals our shared commitment to American scientific leadership.' It's not just about the federal government, either – even state leaders are angling for attention. Mississippi Governor Tate Reeves, who is publicizing his state's AI data center investments and gave a keynote at the Expo told Fortune, 'The leaders in this space are here, and I want to be talking to the leaders that are going to make decisions about where they're making capital investments in the future.' The Expo is hosted by The Special Competitive Studies Project (SCSP), a nonprofit Schmidt cofounded in 2021 and for which he remains the major funder and chair. SCSP operates as a subsidiary of The Eric & Wendy Schmidt Fund for Strategic Innovation, the Schmidt family's private foundation, and is an outgrowth of the now-defunct National Security Commission on Artificial Intelligence — the temporary federal advisory body Schmidt also led from 2018 to 2021. The Expo is about building a community that brings private sector, academia and government into one place, said Ylli Bajraktari, president and CEO of SCSP and Schmidt's co-founder, who previously served as chief of staff to National Security Advisor H.R. McMaster and joined the Department of Defense in 2010. 'Washington is not a tech city, but yet, this is a city where [tech and AI] policies are being developed,' he said. Still, all of this seems really about Schmidt's vision for the future of AI, which he shared in depth in a highly-publicized TED talk last month. In it, he argued that humans should welcome the advancement of AI because our society will not have enough humans to be productive in the future–while delivering a dire warning about how the race for AI dominance could go wrong as the technology becomes a geopolitically destabilizing force. He repeated his hypothetical doomsday scenario in his Expo keynote. Schmidt posits an AI competitor who is advancing quickly, and is only about six months behind the U.S. in developing cutting-edge AI. In other international competitions, this could mean a relatively stable balance of power. But the fear is that once a certain level of AI capability is reached, a steep acceleration curve means the other side will never catch up. According to Schmidt, the other side would have to consider bombing their opponent's data centers to stop them from becoming permanently dominant. It's a scenario – with a proposed doctrine called Mutual AI Malfunction that would slow down each side and control progress – that Schmidt introduced alongside co-authors Henry Kissinger and Daniel Huttenlocher in their 2021 book The Age of AI and Our Human Future. Schmidt shared a thought exercise about what he would do if he could build the U.S. military completely from scratch – no Pentagon, no bureaucracy, no old, obsolete technology – that would basically resemble a tech company: agile, software-driven, and centered around networked, AI-powered systems. 'I would have two layers of drones,' he said. 'I'd have a set of ISR drones [unmanned aerial vehicles used for intelligence, surveillance, and reconnaissance missions]. Those are ones that are high and they have deep looking cameras, and they watch everything. They're connected by an AI network, which is presumably designed to be un-jammable. And then I would have essentially bomber drones of one kind or another, the ISR drones would observe, and they would immediately dispatch the bomber.' With that kind of a defensive system, he added, 'it would be essentially impossible to invade a country by land,' and said that he wanted US assets to be protected by defensive 'drone swarms,' adding that 'there's an entire set of companies in America that want to build this…many of them are here at our show – I want a small amount of the government's money to go into that industry.' Schmidt's Expo is open to all, and there were those in attendance who would beg to differ with his gung-ho takes on the future of battle. The International Committee of the Red Cross, for example, showcased a booth with thought-provoking, graffiti-style questions like 'Does AI make wars better or worse?' Even student attendees well-versed in wargaming, like Luke Miller, a rising sophomore studying international relations at the College of William and Mary and a member of a wargames club, said that today's era of AI and national security is 'supposed to be a sobering moment.' For a country like Ukraine to deploy drones to attack Russian air bases — as they did with great effect just days before the conference— 'is something we should definitely be concerned about going forward,' he said. Still, it was Schmidt's vision of the future of warfare and national security that was front and center at an event 'designed to strengthen U.S. and allied competitiveness in critical technologies.' 'Have you all had a chance to go hang out at the drone cage?' he asked the audience, pointing out the young age of many of the competitors. '[They are] beating the much bigger adults,' he said. 'This is the future. They're inventing it with or without us, that's where we should go.' This story was originally featured on
