Alibaba Unveils Largest AI Model Yet as DeepSeek Rolls Out Ultra-Cheap Alternative

Alibaba unveiled its most powerful AI model to date on Monday, sending its Hong Kong listed shares up 7%. The new model, Qwen3.8-Max, carries 2.4 trillion parameters, placing it close in scale to rival Moonshot AI’s Kimi K3, which has 2.8 trillion parameters. Both models can process text, images and video, handling up to one million tokens in a single request.

Model Rankings Show Strong but Not Top Performance

Qwen3.8-Max quickly became the top-ranked Chinese model on the crowdsourced Arena.AI platform for text-based tasks, though it still trails Anthropic’s Claude Fable 5 and several Opus variants. On the visual analysis leaderboard, it ranked second globally, again behind a Claude Fable 5 variant. Alibaba said the model uses a mixture-of-experts design that activates only a portion of its total parameters per task, helping control costs and response times.

DeepSeek Undercuts Rivals on Price

Separately, DeepSeek released its V4-Flash model on Friday, which research firm Artificial Analysis found to be the cheapest among major AI models to run on benchmark tests. The model costs a fraction of competitors per test, significantly less than Kimi K3, OpenAI’s GPT-5.6 Sol, and Claude Fable 5, based on real-world usage rather than headline pricing alone.

Chinese Firms Bet on Open-Weight Models

Analysts say Chinese AI companies are targeting businesses that need capable, affordable and accessible models rather than only the most advanced systems, a strategy reflected in both companies’ decision to release open-weight versions of their latest models.