Alibaba desafía a OpenAI y Anthropic con su modelo de IA más grande hasta la fecha

El nuevo modelo, con 2,4 billones de parámetros, supera al recién lanzado Kimi K3 de Moonshot en múltiples puntos de referencia, marcando el regreso de la compañía a sus raíces de código abierto tras meses de lanzamientos propietarios. Disponible ya en las API de Model Studio de Alibaba Cloud y en QwenWork, la plataforma de asistente de IA para empresas, el proveedor planea hacer pública la próxima semana una versión que permita a los usuarios ajustar el modelo. Este lanzamiento sitúa a Alibaba en feroz competencia con rivales como OpenAI y Anthropic. A diferencia de los sistemas propietarios de IA que solo ofrecen acceso por API, una liberación de pesos abiertos permite a desarrolladores y empresas descargar los pesos del modelo para personalizarlos y adaptarlos a sus propias necesidades.

Qwen3.8-Max excels at processing and autonomous, says Alibaba

According to Alibaba, the new model achieved results comparable to—or even better than—Anthropic’s Fable 5 in several benchmarks. It can map complex data, such as video streams and thick manuals, into searchable archives. 

Moreover, it recreates apps using screenshots, designs interactive games and educational content, and converts flat floor plans into immersive 3D visualizations. The company also noted that the model excels at autonomous coding, complex research, and other tasks that require extended processing. It proved its autonomy by driving an in-house software project solo for 16 days. Additionally, the latest system boosts efficiency by activating only the required parts, helping reduce processing costs and latency. 

So far, in user-voted arenas, Alibaba’s model has captured top spots, trailing only a few premium Claude models. It secured fifth place in text and second in vision. It initially dropped on July 19 via Alibaba’s subscription plan and Qoder ecosystem. At the time, Alibaba asserted that the system performance was second only to Fable 5.

The debut lifted Alibaba’s shares by as much as 7.3%, their best performance in nearly a month. The market reaction suggests investors view AI as one of Alibaba’s key long-term growth drivers. The company has invested heavily in cloud computing and generative AI, with stronger model performance expected to attract more enterprise customers to its cloud services and developer ecosystem.

China and US AI rivalry continues

The latest launch reflects the increasingly intense competition between Chinese and U.S. AI developers, who are racing to release more capable foundation models at a faster pace. Rather than competing solely on benchmark scores, companies are now differentiating themselves through coding performance, multimodal capabilities, enterprise adoption, and the availability of open-source models that developers can freely customize.

The latest Qwen debut adds momentum to China’s rapid catch-up, arriving days after Moonshot’s Kimi stunned Western tech hubs. Chinese labs are still fighting to eclipse OpenAI and Anthropic using fractionally fewer computing resources. For starters, DeepSeek just expanded its V4 Flash access while ByteDance and MiniMax launched video models on Friday.  

Vey-Sern Ling, managing director at Union Bancaire Privée, even argued that the Chinese models are growing increasingly close in performance with other US superior models. He contended, “Many investors continue to underestimate Chinese AI models because of US chip restrictions or general skepticism. In reality, the gap is probably much closer and narrowing fast. Alibaba’s Qwen 3.8 is another proof point, following Kimi K3.”

Though over the last few months, the rivalry between the two countries has intensified. Earlier, the US government blacklisted foreign access to Anthropic’s Fable 5 and Mythos 5.

The government temporarily restricted the models as a precaution to prevent security bypassing by Chinese authorities. Nonetheless, on Washington’s order, Anthropic returned access to its overseas base. It noted, however, that it was instituting stronger safeguards to protect its technology and had made commitments to the government.

More recently, the firm also accused Chinese developers of distillation attacks. The San Francisco-based company told two U.S. lawmakers in a letter that operators associated with Alibaba used thousands of fake accounts to conduct nearly 29 million exchanges with Claude, calling it the biggest extraction campaign it has seen. The AI startup even urged lawmakers to hold companies behind these attacks accountable and reinforce measures to prevent U.S. technology from being stolen. 

The exploit was highly selective, focusing on Claude’s knack for managing intricate projects and high-level reasoning, according to the San Francisco-based company. It added, “Distillation attacks turn hundreds of billions of dollars in American investment and [research and development] into a massive subsidy for our geopolitical competitors.” 

The smartest crypto minds already read our newsletter. Want in? Join them.