DeepSeek just made the AI pricing war a lot more uncomfortable for everyone else at the table. The Hangzhou-based startup unveiled its V4.1 Flash model on September 10, introducing a 552 billion parameter system that promises to outperform key rivals while costing dramatically less to run.
The immediate market reaction told the story before any benchmark chart could. Shares in Chinese AI competitors MiniMax and Z.ai dropped more than 8% following the announcement, while Alibaba saw its stock slip over 2%.
What makes V4.1 Flash different
The model uses a mixture-of-experts architecture, which is a fancy way of saying it only activates a small slice of its total parameters for any given task. The result is high performance without the computational overhead of lighting up every neuron at once.
DeepSeek built V4.1 Flash with native multimodal capabilities, meaning it can handle text, images, and other data types natively. The company has specifically optimized the model for coding tasks and what the industry calls “agentic” workflows, where AI systems autonomously plan and execute multi-step processes.
The performance claims are notable, but the pricing is where things get genuinely disruptive. DeepSeek’s prior models were priced at roughly $0.14 per million input tokens and $0.28 for output tokens. Earlier DeepSeek models like the V4-Flash were found to be over 100 times cheaper to run than comparable alternatives, averaging about 3 cents per test compared to $3.15 for Anthropic’s Claude Fable 5.
The V4.1 Flash is expected to outperform Moonshot’s Kimi K3 at a significantly lower price point.
A pattern of calculated disruption
The company’s V3 model was trained on 2,000 Nvidia H800 GPUs at a reported cost of $5.6 million. To put that in perspective, leading US AI labs have spent hundreds of millions, sometimes billions, training their flagship models.
Back in 2025, DeepSeek’s strategy of emphasizing low inference costs and open-access models triggered notable market selloffs across the tech sector. The broader implication was hard to ignore: China’s AI ecosystem was producing world-class models despite operating under US chip export restrictions.
What this means for the AI market
The stock market reactions to the V4.1 Flash launch suggest investors are taking the threat seriously. The 8%-plus declines in MiniMax and Z.ai reflect genuine concern that DeepSeek’s pricing strategy could compress margins across the Chinese AI sector. Even Alibaba, with its vastly more diversified business, felt the tremor.
Reports that DeepSeek is preparing for an IPO on Shanghai’s STAR Market add another dimension to the picture. A public listing would give the company access to significant capital while providing a valuation benchmark that could further reshape how investors think about the AI sector.