I just downloaded it to test it out, but it seems like the US might be losing the open source AI race to China.

Qwen, the Chinese open source AI model family, has already passed 700 million downloads. That is roughly twice the population of the USA. What makes it even more interesting is that those numbers are from Hugging Face, a platform that is actually blocked inside China. So ironically, most of those downloads are coming from international users, not domestic Chinese ones.

In the open source world, Qwen is absolutely dominating right now. By early 2026 the Qwen family had spawned more than 200,000 derivative models, making it the most downloaded open source AI model family on Earth. The latest model, Qwen3, is also outperforming Meta’s Llama and DeepSeek in a lot of benchmark categories.

Another thing some people like about it is the privacy angle. Unlike Google’s Gemini ecosystem, which ties together data across your Gmail and other Google services, or OpenAI, which still uses user interactions to improve and train models, Qwen’s open source models can be run locally if you like and are not inherently designed around building a personal profile about you.

Also, open source models like Qwen seem less obsessed with Western style political correctness, identity language policing, or overly cautious moderation compared to some US models. The main thing likely to be more censored on Qwen is discussion around Chinese politics or criticism of the Chinese government.

And the best thing about it, it's entirely free. No usage limits. No credit card information needed to keep it going.

In short, whether people like it or not, China currently seems to be leading the open source AI space by quite a margin.

If you are interested, you can find it in the App Store under Qwen Studio from Alibaba, and the app runs the latest Qwen3.6 Plus model. If you use this hosted app version instead of running the model locally on your own computer, then of course your prompts and conversations are still going through somebody else’s infrastructure and being logged, just like ChatGPT, Gemini, or Claude.

But if all they really have is an email address to store your chat history, and no verified identity or payment information, the model hardly knows who it is actually talking to. You can also use it without logging in at all, although your chat history will not be saved if you ever need it again.

I have no idea what you are talking about. Some AI thing?

6 minutes ago, emptypockets said:

I have no idea what you are talking about. Some AI thing?

Correct. It's another AI platform. Originated in China. It's open source so you can run it locally and privately, but if you download the app version of it that I mentioned then that is offered by Alibaba.

More information from the About tab inside the Qwen Studio mobile app:

The Qwen team is dedicated to pursuing artificial general intelligence (AGI) and is focused on building generalist models, including large language models and large multimodal models. Our mission is to create safe, responsible, and intelligent models, making AGI accessible to the global community through open-source initiatives. We have developed a diverse portfolio of AGI models, with our latest release being the Qwen3 series of large language models, officially launched on April 29, 2025. The Qwen3 series includes multiple models of various sizes, featuring both dense models and Mixture-of-Experts (MoE) architectures. We have also introduced specialized models such as Qwen3-Coder for enhanced programming capabilities. In the multimodal domain, we have released Qwen-Image, a 20B parameter image generation foundation model, and its advanced editing counterpart, Qwen-Image-Edit, released in August 2025. To enable more sophisticated applications, we have developed the Qwen-Agent framework, leveraging the instruction-following, tool-use, and planning capabilities of our models to create versatile agents.

Qwen Studio serves as the primary platform for users and developers to interact with our models, offering comprehensive functionalities such as image and video understanding, image generation, document processing, web search, and tool utilization. We welcome you to explore Qwen Studio and share your valuable feedback with us.

Never mind Qwen, try Deepseek V4 Pro but you're not running it locally even though it's open source.

Edited May 12May 12 by ukrules

10 minutes ago, ukrules said:

Never mind Qwen, try Deepseek V4 Pro but you're not running it locally even though it's open source.

Deepseek is good too, but as I said, Qwen beats DeepSeek in a lot of benchmark categories. That's also why it has the much larger amount of uptake that it has.

Funny also how the Qwen Studio app isn't available through the US Apple App Store, but you can get it from the Apple App Store in many other countries, including Thailand. I wonder if that's US protectionism at play?

I asked AI on google if there are security risks with it and it said:

Yes, there are significant security risks associated with Qwen including Qwen3 models, including high vulnerability to jailbreaking, potential data exfiltration, and the generation of malicious code. While they are popular, open-weight models, they are often flagged for data security, censorship, and privacy concerns in the US leading to bans on some government devices.

5 minutes ago, Fat is a type of crazy said:

I asked AI on google if there are security risks with it and it said:

Yes, there are significant security risks associated with Qwen including Qwen3 models, including high vulnerability to jailbreaking, potential data exfiltration, and the generation of malicious code. While they are popular, open-weight models, they are often flagged for data security, censorship, and privacy concerns in the US leading to bans on some government devices.

Thanks. I just asked Claude "Are there security risks with Qwen including Qwen3 models?", which gave a long and detailed answer, but the summary it gave at the end is:

Bottom line: If you’re using Qwen via Alibaba Cloud/chat.qwen.ai, the data sovereignty concern is real and material. If you’re self-hosting the open weights from a verified source, the geopolitical risk is largely moot, but the jailbreak weakness and the Qwen3 mode-drift issue are worth knowing about for any deployment scenario.

---

By the way, I don't ask ChatGPT important questions like these anymore. It's so often completely off base that I don't even bother anymore. And Grok can be useful, but I don't trust Musk. So I try to rely less on Grok now too. I stick mainly to Gemini and Claude. But this Chinese stuff is very interesting. Especially at their Open Source models run locally. Also, I worry less about privacy and more about accuracy of information. For a consumer level user like me that's the top priority.

Edited May 12May 12 by Terrance8812

If one is a business that is seeking to save costs, then a Chinese model might be more cost-effective, and still meet one's requirements.

But, so far, I am sticking with Gemini, since it's included in my Google Drive subscription.

Gemini is still better than the Chinese models, but perhaps not for long, and so just wait and see.

I would gladly use a Chinese model if I were unable to use Gemini.

Maybe I can try getting the two to talk together, just for fun?

1 hour ago, Terrance8812 said:

I just downloaded it to test it out, but it seems like the US might be losing the open source AI race to China.

Qwen, the Chinese open source AI model family, has already passed 700 million downloads. That is roughly twice the population of the USA. What makes it even more interesting is that those numbers are from Hugging Face, a platform that is actually blocked inside China. So ironically, most of those downloads are coming from international users, not domestic Chinese ones.

In the open source world, Qwen is absolutely dominating right now. By early 2026 the Qwen family had spawned more than 200,000 derivative models, making it the most downloaded open source AI model family on Earth. The latest model, Qwen3, is also outperforming Meta’s Llama and DeepSeek in a lot of benchmark categories.

Another thing some people like about it is the privacy angle. Unlike Google’s Gemini ecosystem, which ties together data across your Gmail and other Google services, or OpenAI, which still uses user interactions to improve and train models, Qwen’s open source models can be run locally if you like and are not inherently designed around building a personal profile about you.

Also, open source models like Qwen seem less obsessed with Western style political correctness, identity language policing, or overly cautious moderation compared to some US models. The main thing likely to be more censored on Qwen is discussion around Chinese politics or criticism of the Chinese government.

And the best thing about it, it's entirely free. No usage limits. No credit card information needed to keep it going.

In short, whether people like it or not, China currently seems to be leading the open source AI space by quite a margin.

If you are interested, you can find it in the App Store under Qwen Studio from Alibaba, and the app runs the latest Qwen3.6 Plus model. If you use this hosted app version instead of running the model locally on your own computer, then of course your prompts and conversations are still going through somebody else’s infrastructure and being logged, just like ChatGPT, Gemini, or Claude.

But if all they really have is an email address to store your chat history, and no verified identity or payment information, the model hardly knows who it is actually talking to. You can also use it without logging in at all, although your chat history will not be saved if you ever need it again.

This sounds like a paid commercial advertisement from the Communist Party of China. Just because someone downloads free software does not make it the best or worthwhile using. I am seriously wary of any software that comes out of China and unwilling to load it on my computers.

13 minutes ago, Screaming said:

This sounds like a paid commercial advertisement from the Communist Party of China. Just because someone downloads free software does not make it the best or worthwhile using. I am seriously wary of any software that comes out of China and unwilling to load it on my computers.

Yep, I got millions from the CCP for posting this topic. Already bought two yachts with the cash and am still trying to decide what to do with the rest of the money. Any suggestions?

As for it being the "best or worthwhile", guess you somehow missed the first sentence of my OP: "I just downloaded it to test it out". So I am not attesting to anything. This is a discussion. An exploration of thoughts and ideas with others who might have tried it, thus the title of the topic: "Have you tried Qwen?".

Get it now?

2 hours ago, Terrance8812 said:

Thanks. I just asked Claude "Are there security risks with Qwen including Qwen3 models?", which gave a long and detailed answer, but the summary it gave at the end is:

Bottom line: If you’re using Qwen via Alibaba Cloud/chat.qwen.ai, the data sovereignty concern is real and material. If you’re self-hosting the open weights from a verified source, the geopolitical risk is largely moot, but the jailbreak weakness and the Qwen3 mode-drift issue are worth knowing about for any deployment scenario.

---

By the way, I don't ask ChatGPT important questions like these anymore. It's so often completely off base that I don't even bother anymore. And Grok can be useful, but I don't trust Musk. So I try to rely less on Grok now too. I stick mainly to Gemini and Claude. But this Chinese stuff is very interesting. Especially at their Open Source models run locally. Also, I worry less about privacy and more about accuracy of information. For a consumer level user like me that's the top priority.

If you need AI to tell you a Chinese App of any type is a security risk you arent living on planet earth

2 hours ago, Fat is a type of crazy said:

I asked AI on google if there are security risks with it and it said:

Yes, there are significant security risks associated with Qwen including Qwen3 models, including high vulnerability to jailbreaking, potential data exfiltration, and the generation of malicious code.

Sounds like more Red Chinese junk. No thanks.

3 minutes ago, Yagoda said:

If you need AI to tell you a Chinese App of any type is a security risk you arent living on planet earth

Thank you for your immaculate wisdom. Your insight is always superbly invaluable. But an opensource Chinese AI model run locally is much different than an AI app. Meanwhile, Qwen is now the most downloaded AI package on the planet. Also, Airbnb, a platform you may have heard of before, used Qwen to train its current AI customer service model. Pinterest is using Qwen too.

Meanwhile, as I mentioned, ChatGPT uses all your chat conversations to train its models. Gemini merges your AI prompts into your Google account with your Gmail, your YouTube comments and viewing history, and your entire Google search history to construct a deep and personal profile of you going on forever. So Gemini is by far the most invasive. And now Claude quietly changed its TOS so that it will retain all your chat data for at least five years too.

So the idea that Chinese AI is the only threat to privacy and security, and that Western AI is the safe option, is complete BS.

2 hours ago, GammaGlobulin said:

If one is a business that is seeking to save costs, then a Chinese model might be more cost-effective, and still meet one's requirements.

But, so far, I am sticking with Gemini, since it's included in my Google Drive subscription.

Gemini is still better than the Chinese models, but perhaps not for long, and so just wait and see.

I would gladly use a Chinese model if I were unable to use Gemini.

Maybe I can try getting the two to talk together, just for fun?

As soon as I see a post containing three AI prompt screenshots, I ignore the entire post. Try conversing normally, with your own thoughts and ideas, like nearly everyone else does...

20 minutes ago, Yagoda said:

If you need AI to tell you a Chinese App of any type is a security risk you arent living on planet earth

What gets me in this day and age knowing that common knowledge is that anything coming out of China is regulated by the CP and their primary intent is to influence as much minds in the world as possible, why would anyone monkey around with their trash. Gullible perhaps?

6 minutes ago, Terrance8812 said:

Chinese AI model run locally is much different than an AI app.

Why would you want to run this locally? Chinese apps are known for data harvesting.

4 minutes ago, Terrance8812 said:

Thank you for your immaculate wisdom.

Your welcome. An authoritarian racist aggressive and revanchist regime that murders 20 million of their own citizens and is noted for surveillance, blackmail and control shouldnt be trusted. Thank you for agreeing.

Whenever there are options, China shouldnt be one

2 minutes ago, novacova said:

Why would you want to run this locally? Chinese apps are known for data harvesting.

It is open source. That means it can be beta and security tested. That is why I think many people are using it, including many large American companies to build their own AI systems, instead of paying millions to Open AI, Anthropic and Google who are openly harvesting your data.

4 minutes ago, novacova said:

What gets me in this day and age knowing that common knowledge is that anything coming out of China is regulated by the CP and their primary intent is to influence as much minds in the world as possible, why would anyone monkey around with their trash. Gullible perhaps?

Or, who knows, we already know how these authoritarian regimes push propaganda. Chinese cars are the best!

6 minutes ago, Yagoda said:

Your welcome. An authoritarian racist aggressive and revanchist regime that murders 20 million of their own citizens and is noted for surveillance, blackmail and control shouldnt be trusted. Thank you for agreeing.

Whenever there are options, China shouldnt be one

You know nothing about technology so you switch to speaking in broad strokes. Fine. So you are living in a country that would be a completely failed state if it weren't dominated by Chinese influence, money, and a Chinese run scam industry that is openly stealing billions from Americans and with that stolen money contributing to more than 50% of your country's GDP. I would take you more seriously if you weren't speaking out of both sides of your mouth at once.

Edited May 12May 12 by Terrance8812

Chinese Qwen Stefani?

I'd definitely try that. But while wearing a facemask.

2 minutes ago, Terrance8812 said:

You know nothing about technology so you switch to speaking in broad strokes.

You dont know what I know or dont know.

3 minutes ago, Terrance8812 said:

So you are living in a country that would be a completely failed state if it weren't dominated by Chinese influnece, money, and a Chinese run scam industry stealing billions from Americans and that contributes to more than 50% of your country's GDP

Got it. Got a source for your figure of 50%, not that your sentence makes any sense.

Methinks we found one here. What country ya from non native English speaker? Syntax matters

Just now, Packer said:

Chinese Qwen Stefani?

I'd definitely try that. But while wearing a facemask.

Yeah, it is their latest AI love doll. I have both models, with and without the mask. Still trying to decide which one I like better. Maybe no need to choose. 3 is a party, right?

8 minutes ago, Yagoda said:

You dont know what I know or dont know.

Got it. Got a source for your figure of 50%, not that your sentence makes any sense.

Methinks we found one here. What country ya from non native English speaker? Syntax matters

Source - Gemini:

According to a landmark May 2024 report by the United States Institute of Peace (USIP), Cambodia’s online scam industry generates an estimated $12.5 billion to $15 billion annually, a figure that is equivalent to approximately half of the country's formal Gross Domestic Product (GDP). This industrial-scale criminal ecosystem, which employs an estimated 100,000 trafficked workers, primarily utilizes "pig-butchering" social engineering schemes and sophisticated cryptocurrency laundering networks. The UN Office on Drugs and Crime (UNODC) further highlights that these illicit revenues across the Mekong region (Cambodia, Laos, and Myanmar) exceed $43.8 billion a year—nearly 40% of their combined formal economies—often operating with the alleged complicity of local elites and through repurposed infrastructure like casinos and Special Economic Zones.

https://www.usip.org/publications/2024/05/transnational-crime-southeast-asia-growing-threat-global-peace-and-security

https://www.unodc.org/roseap/uploads/documents/Publications/2025/Inflection_Point_2025.pdf

21 minutes ago, Terrance8812 said:

It is open source. That means it can be beta and security tested. That is why I think many people are using it, including many large American companies to build their own AI systems, instead of paying millions to Open AI, Anthropic and Google who are openly harvesting your data.

Sure it can be tested for what is currently active at the time of being tested. Code can be programmed to be active only for a few seconds, inserted activated and self delete, then once discovered it’s too late and there’s no telling what fell into the vulnerability trap. Best to use stuff that is sandboxed where nothing gets through to the data source or in/out of the app because the data sits on the other side and not on the device. App developers are information hungry and will go to any extent to get as much information and influence as many as possible. Nothing is free, if it says it is then you’re most certainly handing something over.

3 minutes ago, Terrance8812 said:

Source - Gemini:

According to a landmark May 2024 report by the United States Institute of Peace (USIP), Cambodia’s online scam industry generates an estimated $12.5 billion to $15 billion annually, a figure that is equivalent to approximately half of the country's formal Gross Domestic Product (GDP). This industrial-scale criminal ecosystem, which employs an estimated 100,000 trafficked workers, primarily utilizes "pig-butchering" social engineering schemes and sophisticated cryptocurrency laundering networks. The UN Office on Drugs and Crime (UNODC) further highlights that these illicit revenues across the Mekong region (Cambodia, Laos, and Myanmar) exceed $43.8 billion a year—nearly 40% of their combined formal economies—often operating with the alleged complicity of local elites and through repurposed infrastructure like casinos and Special Economic Zones.

https://www.usip.org/publications/2024/05/transnational-crime-southeast-asia-growing-threat-global-peace-and-security

https://www.unodc.org/roseap/uploads/documents/Publications/2025/Inflection_Point_2025.pdf

Bro what does Cambodia have to do with anything? Im American LOL. Where you from?

3 minutes ago, novacova said:

Sure it can be tested for what is currently active at the time of being tested. Code can be programmed to be active only for a few seconds, inserted activated and self delete, then once discovered it’s too late and there’s no telling what fell into the vulnerability trap. Best to use stuff that is sandboxed where nothing gets through to the data source or in/out of the app because the data sits on the other side and not on the device. App developers are information hungry and will go to any extent to get as much information and influence as many as possible. Nothing is free, if it says it is then you’re most certainly handing something over.

Fair enough. But for regular consumer-level users like me, does it really matter that much if I’m using a chatbot app running on the Qwen platform to get information that isn’t private, especially when I’m mainly interested in better accuracy and less censorship of questions and answers, which is often more restrictive with Western-based AI LLMs?

Also, we already know the extent to which Western AI models like ChatGPT, Gemini, and Claude harvest information, so is it really any different from what the Chinese might secretly be doing with our prompts?

If I were a large corporation with a lot of intellectual property to protect, I might be more concerned. But having said that, many large American corporations are already building AI business models around Qwen.

I’m not saying it’s the ideal solution, but should everyone be so quick to dismiss it, especially given its open-source element?

1 hour ago, Yagoda said:

Bro what does Cambodia have to do with anything? Im American LOL. Where you from?

Says the charmer who saves up his social security checks all year to throw birthday parties for himself at Nana Plaza and lives in Camholio because he can't afford to live anywhere else. 🤣