Alibaba Drops Biggest AI Model as DeepSeek Cuts Costs

Alibaba's Qwen3.8-Max, scheduled for release next week, has 2.4 trillion parameters and supports a context window of up to 1 million tokens, allowing it to process thousands of pages of information. The company described it as one of the most powerful models in its Qwen family.

DeepSeek's V4-Flash, meanwhile, charges $0.14 per million input tokens and $0.28 per million output tokens, according to research firm Artificial Analysis. The firm scored the model 50 out of 100 on its Intelligence Index, which combines results from nine benchmarks covering coding, reasoning, and workplace-style tasks.

Both Qwen3.8-Max and V4-Flash are open-weight models, meaning developers can download the underlying learned settings to run or adapt the systems. This contrasts with closed-source offerings from OpenAI, Anthropic, and Google.

The moves come as Chinese AI startups and tech giants compete fiercely. Last month, Moonshot released Kimi K3, a 2.8-trillion-parameter open-weight model built on a mixture-of-experts architecture, which the company said cuts inference costs compared with similarly sized dense models.

DeepSeek, which shook the industry last year with a high-performing model, now faces domestic rivals including Moonshot, MiniMax, Z.AI, ByteDance, and Alibaba. Reports say DeepSeek is preparing for a potential IPO. Meanwhile, U.S. lawmakers are reportedly weighing how to limit American companies' adoption of Chinese AI models.