Alibaba uncages Qwen-Max as DeepSeek V4-Flash lands at $0.14 per million tokens

The gloves are off in China’s AI model race. Alibaba just released Qwen-Max—its most capable model—as open source, ending its API-only status. This isn’t a hobbyist release; Qwen-Max has been Alibaba’s internal heavyweight, and now anyone can run it, modify it, or build on it directly.

The timing is brutal for Western labs still charging premium API rates. While OpenAI and Anthropic price frontier models around $15-20 per million input tokens, DeepSeek’s newly launched V4-Flash costs $0.14 per million. That’s two orders of magnitude cheaper. And it’s not a toy—it’s a genuinely competitive model that lands squarely in the “cheap and cheerful” category Chinese labs have been perfecting. Together, these moves signal a deliberate strategy: flood the market with capable, open-weight models and collapse the pricing floor.

For US model makers, this creates an existential pricing problem. You can argue about benchmark superiority all day, but when a model is 100x cheaper and open source, enterprise procurement teams start asking uncomfortable questions. Why pay Anthropic $15 when DeepSeek charges pocket change? Why lock into a proprietary API when Qwen-Max runs on your own hardware? The value proposition gets shaky fast.

This is the same playbook that worked for open source software against Microsoft two decades ago—commoditize the complement. By making powerful models dirt cheap and freely available, Chinese labs are trying to do to AI what Linux did to proprietary operating systems. Whether US firms can differentiate on something other than raw model access is now the multi-billion-dollar question.