- Alibaba unveiled its most capable AI model to date, the Qwen3.8-Max, which is not far behind Moonshot's Kimi K3 in size.
- The Qwen3.8-Max has 2.4 trillion parameters, while Moonshot's Kimi K3 has 2.8 trillion parameters.
- Qwen3.8-Max was unveiled on Arena.AI, where it became the highest-ranking Chinese model for text models.
- In terms of image analysis, Qwen3.8-Max ranked second globally on Arena.AI's leaderboard.
- Alibaba stated that Qwen3.8-Max can process up to 1 million tokens at a time.
- Qwen3.8-Max utilizes a mixture-of-experts design, activating only 95 billion parameters at a time to reduce costs.
- Qwen3.8-Max is set to be released next week through Alibaba Cloud's Model Studio platform.
Alibaba's Qwen3.8-Max, unveiled on Monday, is touted as the company's most capable AI model to date, featuring 2.4 trillion parameters. This positions it just behind Moonshot's Kimi K3, which boasts 2.8 trillion parameters. While a higher parameter count does not guarantee superior performance, it remains a critical metric in evaluating AI capabilities.1234567
Qwen3.8-Max was showcased on Arena.AI, where it quickly became the highest-ranking Chinese model for text processing, although it still trails behind Claude Fable 5 and three variants of Opus from Anthropic. In the realm of image analysis, it secured a commendable second place globally, only surpassed by a variant of Claude Fable 5.
Both Qwen3.8-Max and Kimi K3 are designed to handle text, images, and video, with the capability to process up to 1 million tokens simultaneously. Alibaba's innovative mixture-of-experts architecture allows the model to activate only 95 billion parameters at a time, optimizing performance and reducing operational costs.
The model's efficiency was demonstrated when it completed a software-engineering project in just 16 days. Qwen3.8-Max is set to be released next week via Alibaba Cloud's Model Studio platform, as the competition among Chinese tech firms intensifies to develop powerful yet cost-effective AI systems.
Chinese companies are eager to publish parameter counts to attract developers, contrasting with the closed-source models of OpenAI, Anthropic, and Google, which do not disclose such information.
“Qwen3.8-Max packs 2.4 trillion parameters versus Kimi K3's 2.8 trillion, but only 95 billion are active per request under a mixture-of-experts design. It debuted at the top of Arena.AI's Chinese text-model rankings and is due for release next week through Alibaba Cloud's Model Studio.”
