# Alibaba lança modelo de IA que afirma superar o DeepSeek - Reuters **Título Original:** **Alibaba releases AI model it says surpasses DeepSeek** **Fonte**: Reuters **Data de Publicação:** 29.01.2025 **URL:** https://www.reuters.com/technology/artificial-intelligence/alibaba-releases-ai-model-it-claims-surpasses-deepseek-v3-2025-01-29/ **Autor:** Eduardo Baptista # Resumo gerado por AI A Alibaba lançou uma nova versão de seu modelo de inteligência artificial Qwen 2.5, alegando que ele supera o DeepSeek-V3. O lançamento inesperado, durante o feriado do Ano Novo Lunar, sugere pressão da ascensão do DeepSeek nas últimas semanas. A Alibaba afirma que o Qwen 2.5-Max supera o GPT-4o, DeepSeek-V3 e Llama-3.1-405B. - A Alibaba lançou o Qwen 2.5-Max, um novo modelo de IA que afirma superar o DeepSeek-V3.\ - O lançamento ocorreu no primeiro dia do Ano Novo Lunar, indicando pressão da concorrência.\ - A Alibaba afirma que o Qwen 2.5-Max supera o GPT-4o, DeepSeek-V3 e Llama-3.1-405B.\ - O DeepSeek-V3 e seu assistente de IA causaram impacto no Vale do Silício devido aos seus baixos custos.\ - A DeepSeek também lançou o modelo R1, que rivaliza com o o1 da OpenAI.\ - Outras empresas chinesas, como ByteDance, também lançaram atualizações de seus modelos de IA.\ - O DeepSeek-V2 desencadeou uma guerra de preços de modelos de IA na China.\ - O fundador da DeepSeek, Liang Wenfeng, afirma que o objetivo principal é alcançar a AGI.\ - A DeepSeek opera como um laboratório de pesquisa, com foco em inovação. # Artigo Original  BEIJING, Jan 29 (Reuters) - Chinese tech company Alibaba [(9988.HK), opens new tab](https://www.reuters.com/markets/companies/9988.HK) on Wednesday released a new version of its [Qwen 2.5 artificial intelligence model](https://www.reuters.com/technology/alibaba-accelerates-ai-push-by-releasing-new-open-source-models-text-to-video-2024-09-19/) that it claimed surpassed the highly-acclaimed DeepSeek-V3. The unusual timing of the Qwen 2.5-Max's release, on the first day of the Lunar New Year when most Chinese people are off work and with their families, points to the pressure Chinese AI startup [DeepSeek's meteoric rise](https://www.reuters.com/technology/tech-stock-selloff-deepens-deepseek-triggers-ai-rethink-2025-01-28/) in the past three weeks has placed on not just overseas rivals, but also its domestic competition. "Qwen 2.5-Max outperforms ... almost across the board GPT-4o, DeepSeek-V3 and Llama-3.1-405B," Alibaba's cloud unit said in an announcement posted on its official WeChat account, referring to OpenAI and Meta's most advanced open-source AI models. The Jan. 10 release of DeepSeek's AI assistant, powered by the DeepSeek-V3 model, as well as the Jan. 20 release of its R1 model, has shocked Silicon Valley and caused tech shares to plunge, with the Chinese startup's [purportedly low development and usage costs](https://www.reuters.com/technology/artificial-intelligence/american-ai-firms-try-poke-holes-disruptive-deepseek-2025-01-28/) prompting investors to question huge spending plans by leading AI firms in the United States. But DeepSeek's success has also led to a scramble among its domestic competitors to upgrade their own AI models. Two days after the release of DeepSeek-R1, TikTok owner ByteDance [released an update to its flagship AI model](https://www.reuters.com/technology/artificial-intelligence/tiktok-owner-bytedance-deepseek-lead-chinese-push-ai-reasoning-2025-01-22/), which it claimed outperformed Microsoft-backed OpenAI's o1 in AIME, a benchmark test that measures how well AI models understand and respond to complex instructions. This echoed DeepSeek's claim that its R1 model rivalled OpenAI's o1 on several performance benchmarks. ## DEEPSEEK VERSUS DOMESTIC COMPETITORS ![[Clippings/publish/attachments/GVHN3TRO5ZPMLCO5XG6ELFQC7E.jpg|Alibaba Group sign is seen at the World Artificial Intelligence Conference (WAIC) in Shanghai]] The predecessor of DeepSeek's V3 model, DeepSeek-V2, triggered an [AI model price war](https://www.reuters.com/technology/iflytek-enters-chinas-ai-language-model-price-war-2024-05-22/) in China after it was released last May. The fact that DeepSeek-V2 was open-source and unprecedentedly cheap, only 1 yuan ($0.14) per 1 million tokens - or units of data processed by the AI model - led to Alibaba's cloud unit [announcing price cuts](https://www.reuters.com/technology/chinese-tech-giants-slash-prices-language-models-used-power-ai-chatbots-2024-05-21/) of up to 97% on a range of models. Other Chinese tech companies followed suit, including Baidu [(9888.HK), opens new tab](https://www.reuters.com/markets/companies/9888.HK) , which released [China's first equivalent to ChatGPT](https://www.reuters.com/technology/chinese-search-giant-baidu-introduces-ernie-bot-2023-03-16/) in March 2023, and the country's most valuable internet company Tencent [(0700.HK), opens new tab](https://www.reuters.com/markets/companies/0700.HK) . Liang Wenfeng, [DeepSeek's enigmatic founder](https://www.reuters.com/technology/deepseek-founder-liang-wenfeng-puts-focus-chinese-innovation-2025-01-28/), said in a rare interview with Chinese media outlet Waves in July that the startup "did not care" about price wars and that achieving AGI (artificial general intelligence) was its main goal. OpenAI defines AGI as autonomous systems that surpass humans in most economically valuable tasks. While large Chinese tech companies like Alibaba have hundreds of thousands of employees, DeepSeek operates like a research lab, staffed mainly by young graduates and doctorate students from top Chinese universities. Liang said in his July interview that he believed China's largest tech companies might not be well suited to the future of the AI industry, contrasting their high costs and top-down structures with DeepSeek's lean operation and loose management style. "Large foundational models require continued innovation, tech giants' capabilities have their limits," he said. The Reuters Daily Briefing newsletter provides all the news you need to start your day. Sign up [here.](https://www.reuters.com/newsletters/daily-briefing/?location=article-paragraph) Reporting by Eduardo Baptista; Editing by Christian Schmollinger Our Standards: [The Thomson Reuters Trust Principles., opens new tab](https://www.thomsonreuters.com/en/about-us/trust-principles.html)