Two Chinese AI labs have slashed the cost of building and running top-tier artificial intelligence models by 99%, putting advanced AI within reach of far more developers and businesses. DeepSeek and Xiaomi achieved the reduction, making their frontier models cost a fraction of American counterparts like GPT-5.5 and Claude Opus.
Who's behind the price drop
DeepSeek and Xiaomi are the labs that pulled it off. Both are Chinese outfits, though they operate differently. DeepSeek focuses on foundational research and open-weight models, while Xiaomi is better known for consumer hardware but has been quietly ramping up its AI work. Together they've driven down costs that have long kept frontier AI exclusive to well-funded companies.
The exact methods haven't been detailed publicly, but the result is clear: a 99% reduction. That means a model that once cost $1 million to train now costs around $10,000. For inference—the cost of actually running the model for users—the savings are similarly dramatic.
What a 99% cheaper model means
Frontier models are the most capable AI systems—the ones that reason, code, and generate complex content. American models like GPT-5.5 from OpenAI and Claude Opus from Anthropic currently set the bar. But their price tags put them out of reach for many startups, researchers, and small businesses.
Chinese labs changed that math. A 99% cost cut makes frontier-level AI accessible to almost anyone with a credit card. Small teams can now experiment with powerful models without burning through venture capital. Universities in developing countries can run cutting-edge research without begging for grants. The barrier to entry just collapsed.
That doesn't mean the American models are obsolete. They still hold advantages in certain benchmarks, safety features, and ecosystem. But the gap in capability is narrowing fast, and the gap in cost is now reversed.
The price drop reshapes the AI arms race. For the last two years, the story has been about scale: bigger models, more compute, higher costs. DeepSeek and Xiaomi are betting the next phase is about efficiency. If you can deliver 90% of the performance at 1% of the cost, you win the volume game.
China has been investing heavily in AI infrastructure and talent. This cost reduction gives Chinese labs a potential edge in deploying AI at scale—not just in tech, but in manufacturing, logistics, and government services. American companies may need to respond with their own efficiency breakthroughs or risk losing market share in price-sensitive segments.
Regulators are watching too. Cheaper frontier models mean more actors can develop powerful AI, which raises questions about safety and misuse. But the genie is out of the bottle. Cost reductions like this are hard to reverse.
What comes next
DeepSeek and Xiaomi haven't announced their next moves. The immediate question is whether other Chinese labs will follow suit, and whether American labs can match the price cuts without sacrificing capability. OpenAI and Anthropic are both working on smaller, cheaper versions of their models, but 99% is a steep target.
For now, developers have a new set of options. The frontier just got a lot more crowded—and a lot cheaper. No one knows if the price war will escalate further, but the opening salvo is clear.
