Even before the buzz created by Kimi K3 died down, another AI from China has emerged, aiming for the top. Alibaba's latest model, Qwen3.8-Max, could challenge GPT and Claude.
China is hot on America's heels in the artificial intelligence race. While OpenAI and Anthropic seemed to be pulling ahead with their advanced models, a series of recent Chinese models have shown that such dominance is not a given. Kimi K3, in particular, made significant waves with its performance comparable to the top models from OpenAI and Anthropic. While its impact is still being felt, a colossal model has now arrived from Alibaba. Like Kimi K3, Alibaba's new Qwen3.8-Max model offers performance remarkably close to the most advanced closed-source models, despite its open architecture.
According to information shared by Alibaba, Qwen3.8-Max is a massive model with a total of 2.4 trillion parameters. However, the model does not use all these parameters simultaneously during operation. Instead, it leverages a Mixture of Experts (MoE) architecture that activates approximately 95 billion parameters. This approach significantly reduces both processing costs and latency, while enabling the model to have a much larger capacity.
Alibaba Returns to an Open-Source Strategy
Qwen3.8-Max is noteworthy not only for its technical specifications but also for a shift in Alibaba's strategy. Earlier this year, the company had offered some of its high-end models as closed-source. However, with the new model, it is reverting to an open-source approach. Alibaba announced that it would release the model weights next week, making them available for developers to download and use. Thus, Qwen3.8-Max will be the first Max-class Qwen model to be offered as open-source.
The open-source approach allows developers to adapt the model to their own systems and build new solutions on top of it. Considering the recent success of Chinese companies like DeepSeek and Moonshot with open-source models, Alibaba's decision could have a significant impact on the AI ecosystem.
Supports 1 Million Tokens, Can Complete Tasks Lasting Days
According to information shared by Alibaba, one of Qwen3.8-Max's most striking features is its 1 million token context window. This enables the model to analyze hundreds of pages of documents, long video recordings, or extensive datasets in a single pass. The company also states that the model can create searchable knowledge bases from video content and live streams.
The company positions Qwen3.8-Max as an ideal model especially for autonomous software development, long-term research projects, and tasks requiring AI agents to operate continuously for days. Indeed, according to Alibaba's internal tests, the model successfully completed a software engineering project independently over 16 days. This can be considered a significant step in the "agent-based artificial intelligence" approach that the industry has been heavily focusing on recently.
0 Comments: