Alibaba

The head of AI at Alibaba leaves the company. That points to a 180º turn for the Qwen family models

by

An employee leaving a company does not have to mean a radical change, especially when that employee has been the leader of an important project and his departure occurs just after the launch. This is what just happened with Junyang (Justin) Lin, the technological leader of the team qwen. A strange exit. On March 2, Alibaba launched a new model family lightweight with two fast models designed for edge use, a multimodal model for agentic systems and a reasoning model that stood up to much larger models. The next day, Junyang Lin announced on his X account “I am leaving. Goodbye, my dear Qwen,” without giving further details. And he wasn’t the only one. Also leaving the company were Hui Binyuan, a scientific researcher, and Yu Bowen, head of post-training at Qwen. No one has commented on the reasons behind his departure from the company and rumors that they had been fired They didn’t wait. However, according to Panda Daily, Alibaba said it had approved his resignation. ¿What is happening? Justin’s departure caused a stir among his colleagues, with some claiming that it was “the end of an era”. We are talking about the person who has led the Qwen team from the beginning and a great AI researcher, with an academic profile that exceeds 40,000 citationsso this decision has raised many eyebrows. Whether fired or resigned, Justin was a key figure on the team, but he also leaves just after a launch and several other employees have followed him. What is happening at Alibaba? Closed models. As we said, the parties involved have not offered more details, but the theories have not been long in coming and one of them is that Alibaba could be thinking of moving towards closed models. Alibaba has been making efforts to monetize its AI and closing their models could be part of the plan. It would certainly make sense for the project leader to quit at the prospect of such a profound change. There’s a new guy in the office. Shortly after the news broke, another one jumped out: Alibaba has signed Zhou Haowho until now was a researcher at Google DeepMind. Zhou will join the Qwen team as head of post-training, so he will directly replace Yu Bowen and not Justin. Zhou has been a key figure in the development of Gemini 3, the Seeker’s AI mode, and Deep Research mode. lto open source strategy. DeepSeek, Kimi, Qwen… Chinese companies have become the standard bearers of open source AI, an antagonistic strategy with the closed stance of the US. But it is not a question of giving away AI just for the sake of it, but rather it is part of their roadmap: offering access to create a large user base and thus be able to be dominant in the future. Furthermore, Chinese companies know very well that the US is technologically ahead (Justin himself recognized it recently), so launching open and free AIs is a way to gain ground on them. However, in the long term it does not seem like a very good strategy because there will come a point where they want to monetize it and there is a risk of losing users who feel betrayed. We do not know if Alibaba has already started down this path, but if it has, we will soon see if this risk is real or not. Image | qwen In Xataka | China’s open AIs aren’t “beating” ChatGPT, they’re doing something more important: catapulting their industry

ByteDance, Alibaba and Tencent are spending $647 million on AI. Or rather: in Christmas bribes by AI

by

The big three Chinese tech companies have decided that the best way to get users for their AI chatbots is to literally pay them to use them. Between them, they are investing more than $2.9 billion in incentives during the Lunar New Year, the biggest Chinese holiday. It is a war with a single intention: to be the gateway for AI in the country. Subsidy war. The Chinese Lunar New Year has become another major battleground to win the AI ​​race. As they say from the LatePost newsletter (translated by Recode China AI), Alibaba leads with 3,000 million yuan (about 431 million dollars) that it will distribute to its users for its app qwenfollowed by Tencent with 1 billion yuan to yuanbaoand Baidu with 500 million. ByteDance, for its part, has secured the most expensive sponsorship of the Spring Festival Gala to promote Doubaoits chatbot that already has 100 million daily active users. In Xataka ByteDance is not satisfied with TikTok and has just started a new career: one that leads it to create its own AI chip User acquisition. Companies are using money in different ways but with the same objective: hooking users. Alibaba is subsidizing real purchases, from milk tea to hotel reservations, all through its Qwen assistant. According to Bloombergsome stores that offered milk tea have been overwhelmed by orders that had been placed through the chatbot. Tencent offers digital envelopes of up to 10,000 yuan (1,219 euros) directly in cash. On the other hand, ByteDance has taken advantage of its muscle in social networks to integrate Doubao throughout its network of applications. Between the lines. The most interesting part of all this is that it seems that none of these companies yet know how to monetize their AI tools, according to industry sources cited by LatePost. “Monetization models for Chinese AI companies remain murky, a challenge that is also reflected in the United States,” points out Shi Jialong, analyst at Nomura. They are buying users in the hopes of later figuring out how to convert them into revenue. {“videoId”:”x8jpy2b”,”autoplay”:false,”title”:”What’s BEHIND AIs like CHATGPT, DALL-E or MIDJOURNEY? | ARTIFICIAL INTELLIGENCE”, “tag”:”Webedia-prod”, “duration”:”1173″} Competence. The situation is radically different from that of a year ago. DeepSeek changed the rules of the game your R1 model last year, gaining 10 million active users in less than a month. And just as they mention in LatePost, that set off a chain reaction, causing Tencent to dive headlong into AI after years of caution, Alibaba to prioritize its Qwen app above everything (even its Quark browser), and ByteDance to accelerate its investment in talent and infrastructure. Yields. ByteDance reported net profits of about $40 billion in the first three quarters of the year, while Tencent reached $30 billion and Alibaba about $10 billion. according to LatePost. Despite having achieved lower profitability in its operations, Alibaba intends to increase its investment in AI infrastructure, specifically from 55 billion to 69 billion dollars in the next three years, as pointed out in the newsletter. ByteDance, for its part, was processing an average of 63 billion tokens daily with its AI models at the end of 2025, a growth of 200% in six months. In Xataka "The world is in danger": Anthropic’s security manager leaves the company to write poetry And now what. The subsidy war to be the gateway to China is not new. As well as remember In Bloomberg, in sectors such as shared transportation or food delivery, they have experienced this battle of companies throwing incentives at their users. And companies lose money massively until the market consolidates. The difference is that here users are not afraid to change AI models and quickly switch to the one that offers the best technical performance, as indicated the OpenRouter report. It will be interesting to see what the market share of the main AI models in China looks like when they stop flying the envelope. Cover image | Arthur Wang and Solen Feyissa In Xataka |Google is going to borrow money to pay back in 100 years. You have to believe that in 100 years Google will still be there (function() { window._JS_MODULES = window._JS_MODULES || {}; var headElement = document.getElementsByTagName(‘head’)(0); if (_JS_MODULES.instagram) { var instagramScript = document.createElement(‘script’); instagramScript.src=”https://platform.instagram.com/en_US/embeds.js”; instagramScript.async = true; instagramScript.defer = true; headElement.appendChild(instagramScript); – The news ByteDance, Alibaba and Tencent are spending $647 million on AI. Or rather: in Christmas bribes by AI was originally published in Xataka by Antonio Vallejo .

While OpenAI takes all the media glory with ChatGPT, Alibaba is already taking important clients with Qwen. The latest: Airbnb

by

Alibaba has been investing in its family of open language models for quite some time.qwen‘, which are gaining increasing acceptance between developers and users. Although OpenAI takes all the media glory with ChatGPT and the rest of the services, the Chinese firm is not short and already is overtaking him with some clients. The latest example: Airbnb, which has chosen to rely mostly on Alibaba’s Qwen AI model for its automated customer service, leaving ChatGPT in a secondary role. Airbnb’s decision. Brian Chesky, co-founder and CEO of the tourist accommodation platform, explained Bloomberg this week that his company “heavily relies” on Alibaba’s Qwen model. As he admitted to the outlet, ChatGPT’s integration capabilities “are not quite ready” for Airbnb’s needs. On the other hand, Chesky assured that Qwen is “very good, fast and cheap.” It is curious, especially considering that Chesky is a personal friend of Sam Altman, head of OpenAI. How the system works. Airbnb’s customer service agent, which the company deployed to all its users Americans in English last May, is built on 13 different AI models, including those from OpenAI, Google and open source providers. However, Chesky recognized that, although they use the latest OpenAI models, “we usually don’t use them much in production because there are faster and cheaper models.” Just like point the company, the system has allowed them to cut their human workforce by 15% and claims to have saved average resolution time, going from almost three hours to just six seconds. Open source is gaining ground. Open source models, which developers can modify as they wish, are increasingly challenging closed systems like those from OpenAI. Although the company also has an open model (gpt-oss), Chinese tech companies are releasing models much faster, more cost-effectively, and open source. Joe Tsai, president of Alibaba, declared recently that the winner in AI should be determined by “who can adopt it the fastest,” not “who creates the most powerful model.” A future integration with ChatGPT in the air. Although Airbnb is awaiting the development of ChatGPT app integrations and could consider a collaboration in the future, similar to those of its competitors Booking and Expedia, the platform is not currently among the first applications available on the OpenAI chatbot. Chesky even advised to OpenAI about its new ability for third-party developers to integrate their applications into ChatGPT, a feature that the company announced this month and which he described as a “developer preview.” And now what. Airbnb plans expand its AI agent with support in Spanish and French this fall, and 56 more languages ​​next year. Meanwhile, the company claims to be betting on new social functions to foster connections between users and improve travel recommendations within the application. For Chesky, these features are “probably the most differentiated part of Airbnb.” Cover image | Unsplash (Oberon Copeland), Wikimedia In Xataka | OpenAI is no longer a startup. Now it is a black hole of 500,000 million that threatens the world economy

Hangzhou is the city of DeepSeek, Alibaba and Unitree without any of the typical Silicon Valley ingredients. His secret is another

by

Hangzhou, a city of 12 million inhabitants 180 km south of Shanghai, is home to a striking number of powerful technology companies: Seven reference technologies (the six ‘little dragons’ plus the giant Alibaba) in a city that does not have any of the elements considered essential in Silicon Valley: Abundant venture capital. Leading universities. Links between university and industry. And a robust industrial structure. How could you then Hangzhou emerge well? The facts. Venture capital is plummeting in China. Funds in yuan have fallen from 88.42 billion dollars in 2022 to 5.38 billion in 2024. Funds in dollars, from 17.32 billion to 750 million. Hangzhou has not been a major recipient of investment until last year, when its province –Zheijang– stood out with 41 new corporate venture capital funds. But it was only after Unitree or Game Science had gained national attention. Missing. Hangzhou has only one elite university – Zhejiang – compared to 26 in Beijing, 11 in Jiangsu or 10 in Shanghai. The admission rate at Tsinghua and Beijing Universities for students from the capital (0.85%) is almost ten times that of students from Zhejiang (0.09%). None of the founders of “the six little dragons” or Alibaba created their company directly from university. Liang Wenfeng founded High-Flyer, the hedge fund after DeepSeekeight years after graduating. Jack Ma was rejected for 30 jobs after finishing his studies. Yes, but. The city has innovated by doing away with those ingredients. The explanation offered by Zilan Qian, a researcher at the Oxford China Policy Lab, points out ChinaTalk to “flexible governance”: a model where officials adopt “waiters” and “babysitters” mentality that facilitate rather than control. The context. Hangzhou does not have the political, financial or industrial importance of first-tier cities, which has given it greater local autonomy to shape its technology sector. Zhejiang province was a pioneer since the 1980s in promoting private enterprise during the early phases of Chinese economic reforms. Jack Ma He tried to establish Alibaba’s headquarters in Beijing or Shanghai, but failed due to the cost of rent and bureaucratic barriers. In 2015, Ma explained her decision: “Beijing favors state-owned enterprises, Shanghai favors foreign companies, and Alibaba was nothing in their eyes. If we return to Hangzhou, we become the local only child who receives all the attention and support.” Hangzhou is part of the sometimes called “chinese technology triangle“(sometimes also”golden triangle“) along with Shanghai and Shenzhen. More than a geometric reality, the functional metaphor describes the complementarity of three cities: Shenzhen provides industrial capacity and hardware. Shanghai concentrates finances and internationalization Hangzhou stands out in the internet, AI and an ecosystem favorable to private companies. Each vertex of the triangle has different strengths that, combined, generate an ecosystem where geographical proximity facilitates collaboration and flow of talent between the three poles. Between the lines. The model is described as “market-oriented” but maintains a level of centralized governance. The Hangzhou government sees quality of life as a strategy to attract businesses and talent, but positions itself as an enabler, not a controller. The absence of state-backed research institutes and a strong industrial base contributes to the government’s humble attitude. If Hangzhou were more strategic or more industrial, DeepSeek might not have had the creative space to emerge and provoke the earthquake that caused in January. The narrative of “self-made industry” and “entrepreneurial bureaucracy” admits conflicting readings. What some interpret as facilitation, others read as a euphemism for “dirigiste intervention by the State”, with a very defined plan of action and long-term objectives. “Flexible governance” can be both real local autonomy… and dirigisme disguised as pragmatism. At least it is no longer “a city south of Shanghai” but “Alibaba City” or “DeepSeek City”. In Xataka | China is selling us a future full of humanoid robots. We have (many) doubts Featured image | JinHui CHEN in Unsplash

Jack Ma, founder of Alibaba, is a Tai Chi fan and millionaire. That’s why he gave himself the gift of starring in a martial arts movie

by

Jack Ma, founder and executive chairman of Alibaba, is known worldwide for having founded a technology empire in China, but he also made his first steps in the world of cinema. And of all genres, martial arts. An eccentricity that was allowed to be filled with top stars of the genre and that, however, was more than just an idea to massage the ego: Ma has been practicing Tai Chi for 30 years. The man, the legend. Jack Ma, born in 1964 in Hangzhou, China, is the founder in 1999 and former CEO of Alibaba Group, one of the largest e-commerce conglomerates in the world. He founded the company with the goal of connecting Chinese companies with international buyers, revolutionizing digital commerce. His vision turned Alibaba into a global e-commerce and technology giant. He is currently dedicated to philanthropy and education, while his company has been facing a fall of influence very considerable. 22 minutes of tollinas. In 2017, before retiring from Alibaba, Jack Ma starred in and co-produced a martial arts short film titled ‘Gong Shou Daoo’ (literally, “The Art of Attacking and Defending”): 22 minutes depicting Ma as a Tai Chi master facing off against a series of rivals, all of whom are legendary actors in Chinese action cinema. People like Jet Li, Donnie Yen, Tony Jaa, Wu Jing, Natasha Liu Bordizzo and boxing champion Zou Shiming, among others. Everyone involved agreed to participate free of charge, motivated by the promotion of Chinese culture and martial arts globally. All for Tai Chi. Jack Ma, as we have mentioned, has been practicing Tai Chi for thirty years, and used the film to highlight the philosophy and values ​​of this martial discipline, focusing on balance and harmony. The project arises from the intention of preserving and disseminating this discipline, and reflecting on the screen Tai Chi as an art with a physical and a spiritual aspect. Result: more than 170 million views when it was launched in 2017. But it’s okay or what. Let’s see, a movie in which Donnie Yen and Tony Jaa share the bill cannot be bad even if it wants to. But also, the fan has the opportunity to meet Jet Li again, who has been semi-retired from action movies for years, in a reasonably extensive role with fights. Ma is neither the most forceful nor the most expressive actor, but for that gift to fans alone, he has our thumbs up. If you want more Tai Chi. Speaking of Jet Li and speaking of Tai Chi, the choreographer of the film is Ku Huen-Chiu, chosen very consciously by Ma. Apart from a rich career with films in his filmography such as a sequel to ‘The Matrix’, ‘Kill Bill’ and several classics of Hong Kong action cinema from the nineties, Ku Huen-Chiu participated in the filming of ‘Tai Chi Master’, a marvel by Li and Michelle Yeoh directed by none other than Yuen-Woo Ping, one of the great Chinese action choreographers of all time. So if you think Tai Chi is something for old people in the park, check out that hilarious dynastic martial epic and you’ll change your mind. In Xataka | Jack Ma was the richest man in China. Until he fell out with the government and stopped being one

Alibaba has one of the best open source AI models. Your next step: use it in robotics

by

Alibaba has taken another step in its commitment to artificial intelligence by creating an internal team dedicated to roboticswhich will operate from qwenits AI modeling division. The Chinese giant, owner of one of the best open source AI models, now wants the Qwen team to know how to apply its knowledge in robotics, a sector that is beginning to awaken interest, not only in industry, but also with the arrival of projects in the domestic sphere. Who leads the project. Justin Lin, technology manager at Qwen and expert in multimodal models (capable of processing text, sound and images), was the one who has confirmed the creation of this “small team for robotics and embodied AI” through their social networks. Lin has worked on the development of the Qwen models, which are currently among the most popular in open source globally. The vision behind the movement. According to Linmultimodal AI models are evolving into “fundamental agents” capable of performing complex long-term reasoning tasks thanks to reinforcement learning. “They should definitely make the leap from the virtual world to the physical world,” he said. explained the manager, making clear the intention to apply these technologies in tangible devices. Alibaba’s big bet. This announcement is part of Alibaba’s broader strategy in the sector. Last month, the company led a financing round of 140 million dollars at X Square Robot, a Chinese robotics startup. In addition, its CEO Eddie Wu esteem that global investment in AI will reach $4 trillion in the next five years, a figure that reflects the sector’s expectations. Global competition. Alibaba is not alone in this race. Nvidia and SoftBank are also making significant moves in smart robotics. SoftBank just announced the acquisition of ABB’s industrial robots business for $5.4 billion, while Nvidia CEO Jensen Huang has qualified the combination of AI and robotics as a “multi-billion dollar” long-term growth opportunity. China is also the world’s leading power in the robotics sector. And only in 2024, Chinese factories installed nearly 300,000 industrial robotsa figure higher than the rest of the world combined. The Qwen factor. The choice to place this team within Qwen makes all the sense in the world. Seven models of the Qwen series are currently listed in the top 10 Hugging Facewith the multimodal model Qwen3-Omni occupying first place. This strength in AI provides the company with a solid foundation to develop advanced robotic applications based on the journey they already have with Qwen. Cover image | zhang hui and Possessed Photography In Xataka | AI companies have just encountered an unexpected challenge: insurers have started to turn their backs on them

The new King of the AI ​​Open Source is Alibaba. And its strategy is simple: to be tired

by

Alibaba qwen3 -omni has become the new jewel of the AI ​​Open Source segment. This model, launched last week by the Chinese giant, manages to compete in various benchmarks with some of the best models of OpenAi or Google. But the important thing is not so much as the fact that it alibba Not stop taking AI models at a frantic pace and almost strenuous. QWEN models succeed. The QWEN3-OMNI-30B-A3B-INSTRUCT model is one of the variants of the QWEN3-OMNI family newly launched by Alibaba. This version has become the most popular model in the Ranking of available models in Hugging Face. Since it appeared there, almost 100,000 times has been downloaded, but it is not alone. The new QWEN3-Max-Thinking manages to match or overcome models such as Grok 4 or GPT-5 Pro. They do not stop launching models. As they point out In SCMPto date Alibaba has published more than 300 Open Source models that have served for other developers and companies to launch their own. In fact, it is estimated that there are more than 170,000 derived models, which seems to have managed to have Alibaba right now the world’s largest ecosystem. The data were shared in The APSARA conference Organized by Alibaba Cloud in Hangzhou last week. Some recent examples of that frantic rhythm: Alibaba Copa the ranking. Although they were released in April, the QWEN3 models have not stopped renewing in recent months with new capacities in the generation of text, images, audio and even video. The improvements in multimodal behavior have helped create this new “OMNI” family – which precisely handles all kinds of entrances and exits – and with it with it returns that yield They even rival with the best proprietary models of firms such as Google and Openai. Models for all tastes. If one looks That rankingfive of the first 10 models are from Alibaba. Tencent has two others, IBM Granite surprises in fourth position and also We have Deepseek-V3.1-terminus Already a voice text model called VoxCPM. Source: The Atom Project. Llama, missing. Meanwhile, the traditional dominator of said scope, The Meta Llama Modelis totally missing from the first positions of this ranking and appears in position 41. OpenAI and its GPT-Oss-20b model It is also quite displaced (position 30). The responsible for The Atom (American Truly Open Models) Project recently revealed A study in which they highlighted how accumulated discharges of Open Source models already come from Chinese models than US models. Llama was the most downloaded open model until recently. Now that position is occupied by the Models of the Qwen family of Alibaba. Source: The Atom Project. Be careful, downloads are something else. It must be said that the ranking focuses on “trending” models, that is, those whose recent popularity is high. The Openai model has in fact downloaded 6.71 million times, while Alibaba’s most downloaded model is QWEN3-Next-80B-A3B-Instruct, with 2.63 million downloads. Llama-3.1-8b-Instruct surpasses both (for now) with 7.18 million. In The Atom Project, yes, they point out that the accumulated discharges of the different flame variants have just fell below those of the Qwen variants. The reason is simple. Alibaba does not stop getting more and more models. Alibaba’s strategy has been overwhelming, and since in April it launched the first QWEN3 models, it has not gone from maintaining a frantic pace of launching of improved and derived versions such as QWEN3-Next, QWEN3-OMNI or QWEN3-MAX, in addition to specific models for generation of images as qwen-image-editordirect competitor of the famous Nano Banana, from Google. In Xataka | There are many “internal” races within the great AI race. And the Open Source is winning Alibaba

Alibaba is becoming the Ai Open Source sponator. Your family of Qwen models is putting the market above

by

The Chinese giant Alibaba has launched Officially QWEN3-OMNI, an open source artificial intelligence model that can process text, images, audio and video simultaneously. In fact, it is the first model that unifies these four modalities natively and does it completely free, something that none of its US competitors offers. Bet on the Free Code. While Openai and Google charge for using their most advanced multimodal models, Alibaba gives theirs under Apache 2.0 license. This means that any company can download it, modify it and use it commercially without any cost. This open source approach It is the trend that multiple Asian giants are adopting to cause global interest in their language models and that multiple developers around the world want to contribute to their evolution. It is part of China’s strategy to remain relevant in the AI ​​career. Image: Alibaba What can you do exactly. As points The company, QWEN3-OMNI simultaneously processes text in 119 languages, recognizes voice in 19 languages ​​and can speak in 10 different languages. Its “thinker-speaker” architecture separates the reasoning of the audio generation, promising real-time responses with latencies of just 234 milliseconds for audio and 547 milliseconds for video. Benchmarks. In 36 reference tests, QWEN3-OMNI exceeds open source models in 32 of them and establishes new general records in 22. In advanced mathematics (Aime25) obtains 65 points compared to 26.7 of GPT-4O. In writing tasks (Writingbench) 82.6 points, exceeding 75.5 GPT-4O points. While it is true that it is not being compared to Openai’s most avant-garde model to date (GPT-5), it is a real achievement what giants like Alibaba are doing with their free and open source models. Strategy. Alibaba is running a risky but intelligent play: democratize the multimodal AI to gain market share. “This could bring some changes to the panorama of the OMNI open source models,” explained The Qwen team. The announcement occurs just when Nvidia announces Investments of 100,000 million dollars in data centers for OpenAI, while Alibaba and the rest of Asian giants prefer to dispute technological leadership in AI from another angle. What does it mean. Great American technology have opted for proprietary models that generate direct income. Alibaba wants to change the rules by giving instant access to its technology to millions of developers. Even if they offer it for free, they are building an ecosystem that gives them competitive advantage In the long term. And now what. China is not the only one that launches free code models. OpenAi has GPT-Oss And Google has Gemma. Two options that developers have on hand to deploy their ideas, modify them, contribute to their evolution and others, although they are not the main approach of both companies. In the case of Alibaba models, Deepseek either Tencentthe idea does revolve around the open source, and the pulse does not tremble when offering their most powerful models for free (despite the fact that some more complete and specific options are reserved for special agreements). QWEN models A great reputation have been carved Throughout these last years, and this new evolution in his family marks a new ribbon for the rest of the companies, not only in efficiency, but in the deployment of this business model. Cover image | Alibaba and Growika In Xataka | Eight people. An hour of work. A budget dollar. 5,000 new podcasts thanks to AI

Alibaba has just demonstrated that Openai spends 78 million to do the same as them for $ 500,000

by

There is a new star technique to train AI models super efficiently. It is at least what Alibaba seems to have demonstrated, that Friday presented His family of QWEN3-next models and did so presuming from spectacular efficiency that even Leave behind the one he achieved Deepseek R1. What happened. Alibaba Cloud, the Alibaba group’s cloud infrastructure division, presented a new generation of LLMS on Friday that described as “the future of efficient LLMs.” According to those responsible, these new models are 13 times smaller than the largest model that that company has launched, and that was presented just a week earlier. You can try QWen3-Next On the Alibaba website (Remember to choose it from the drop -down menu, in the upper left). QWen3-Next. This is what the models of this family are called, among which it stands out especially QWen3-Next-80b-A3Bwhich according to developers is up to 10 times faster than the QWEN3-32B model that was launched in April. The really remarkable thing is that it also manages to be much faster with a 90% reduction in training costs. $ 500,000 is nothing. According to AI Index Report From Stanford University, to train GPT-4 OpenAI invested $ 78 million in computation. Google was further spent on Gemini Ultra, and according to that study the figure amounted to 191 million dollars. It is estimated that QWEN3-Next has only cost $ 500,000 in that training phase. Better than its competitors. According to the benchmarks made By the artificial firm Analysis, QWen3-Next-80B-A3B has managed to overcome both the latest version of Deepseek R1 and Kimi-K2. Alibaba’s new reasoning model is not the best in global terms-GPT-5, Grok 4, Gemini 2.5 Pro Claude 4.1 Opus overcome it-but still achieves outstanding performance taking into account its training cost. How have you done it? Mixture of experts. These models make use of the Mixture of Expert architecture (MOE). With it, the model is “divided” into a kind of neuronal subnets that are the “experts” specialized in data subsets. Alibaba in this case increased the number of “experts”: while Depseek-V3 and Kimi-K2 make use of 256 and 384 experts, QWen3-Next-80b-A3B makes use of 512 experts, but only activates 10 at the same time. Hybrid attention. The key to that efficiency is in the so -called hybrid attention. Current models usually see their efficiency reduced if the input length is very long and have to “pay more attention” and that implies more computing. In Qwen3-Next-80b-A3B, a technique called “Gated Deltanet” is used that They developed and shared MIT and NVIDIA in March. GATED DELTANET. This technique improves the way in which the models pay attention when making certain adjustments to the input data. The technique determines what information retain and which can be discarded. That allows creating a precise and super -efficient cost mechanism. In fact, QWEN3-Next-80B-A3B is comparable to the most powerful Alibaba model, Qwern3-235B-A22B-Thinking-2507. Efficient and small models. The growing costs of training new models of AI begin to be worrisome, and that has made more and more efforts to create “small” language models that are cheaper to train, are more specialized and especially efficient. Last month Tencent presented models below 7,000 million parameters, and another startup called Z.AI published its GLM-4.5 Air model with only 12,000 million active parameters. Meanwhile, large models such as GPT-5 or Claude use many more parameters, which makes the necessary computation to use them much greater. In Xataka | If the question is which of the great technology is winning the AI ​​career, the answer is: None

Alibaba has presented its largest AI model, with a billion parameters. The question is whether at this point that means something

by

The Chinese giant Alibaba has announced a new language model, the largest they have announced to date. It is called Qwen-3-Max and presumes that it has more than 1 billion parameters. The biggest. It is the last model within the series Qwen3 which was launched in May of this year and, as its name ‘Max’ indicates, it is the largest to date. Its size is given by the parameters, 1 billion to be exact, while the previous models of its series reached a maximum of 235,000 million. According to South China Morning Post (Which owner Alibaba), his model stands out in understanding of language, reasoning and text generation. Benchmarks. The results of the benchmarks place QWen3-Max ahead of competitors such as Claude Opus 4, Deepseek v3.1 and Kimi K2. If Gemini 2.5 Pro or GPT-5 does not appear, it is because they are models of reasoning and have only compared rapid response models. As they point out in Dev.toboth Gemini 2.5 Pro and GPT-5 obtain higher scores in mathematics and code, so reasoning models continue to have advantage in those areas. Qwen3-max-preview can already be tested free of charge. Benchmarks shared by Alibaba. Parameters. The parameters are all the internal variables that a model learns during training. In other words, it is the knowledge that the model has obtained from the data with which it has trained and allows it to interpret our requests and generate their answers. In theory, the more parameters, the model will have more and better capabilities. It also implies that it needs more computational power both to train and to execute the model. More does not mean better. The speech of the parameters remembers that of the megapixels with the first cameras. A 100 megapixel sensor will take larger photos than a 10 sensor, but there are other crucial factors that affect image quality such as sensor size or lens luminosity. Quality data. More parameters can be translated into more learning capacity and more resolution of complex tasks, as long as quality training data has been used. It is obvious: a language model that has been trained with redundant, incorrect or biased data will learn and continue to reproduce those errors in their operation. There are more. In 2022, the laboratory Deepmind from Google, discovered that many models were oversized in parameters but underlined in data. To demonstrate it they created the Chinchilla model with “only” 70,000 million parameters, but four times more data. The result was that it beat Gopher, a model with four times more parameters. Architecture. The architecture of the model is another decisive factor in order to achieve an efficient model; A standard architecture is not the same that forces the model to use its entire neuronal network, than one like Mixture of experts which consists of many smaller networks. It would be something like having an expert committee each with a specialty. In this way, the model can choose your expert for each query and not have to use the entire network. For example, with this technique, Mistral manages to use only a fraction of his parameters And so it is faster and cheap to execute. Image | Markus Winkler, via Pexels In Xataka | The ASML-Mistral alliance reveals the European plan B: if we cannot manufacture chips, we will at least control how they are manufactured

Log In

Forgot password?

Forgot password?

Enter your account data and we will send you a link to reset your password.

Your password reset link appears to be invalid or expired.

Log in

Privacy Policy

Add to Collection

No Collections

Here you'll find all collections you've created before.