DeepSeek V4-Pro Shows China’s AI Price War Is Not Slowing

DeepSeek has launched V4-Pro on OpenRouter, and the new model shows that China’s AI price war is not slowing down. If anything, it is becoming more aggressive.

OpenRouter’s side-by-side comparison lists both Kimi K3 and DeepSeek V4-Pro with a 1,048,576-token context window. The pricing difference is the shock. Kimi K3 is listed at $2.80 per million input tokens and $14 per million output tokens, while DeepSeek V4-Pro is listed at $0.435 per million input tokens and $0.87 per million output tokens.

That means DeepSeek is not only trying to match Chinese AI rivals on capability. It is trying to undercut them hard on price. For developers, startups and companies using AI APIs at scale, that difference matters. Model quality is important, but inference cost can decide which model gets used in production.

The Information reported that DeepSeek’s V4-Pro rivals Kimi K3 on some benchmarks while coming in at much lower prices. South China Morning Post framed the update more cautiously, saying the model struggled on some benchmarks but showed strength in cybersecurity. That mixed reception is worth noting because price alone does not make a model a winner. Reliability, tool use, coding performance and enterprise support still matter.

Still, DeepSeek has been here before. The company has repeatedly put pressure on bigger labs by offering strong models at prices that force the rest of the market to respond. We wrote earlier about DeepSeek V4 Flash raising pressure in the AI agent price war, and V4-Pro continues the same playbook at the higher end.

The Kimi comparison is especially important because Kimi K3 became one of the biggest AI stories of the summer. Moonshot’s model created demand pressure, benchmark anxiety and even geopolitical discussion around Chinese AI progress. We covered how Kimi K3 became a broader AI market headache, and DeepSeek’s new pricing now puts direct pressure on that momentum.

This also affects U.S. labs. OpenAI, Anthropic, Google and xAI can still argue that their best models are stronger, safer or better supported. But enterprise customers are not always buying the absolute strongest model. They are buying the model that is good enough, reliable enough and cheap enough for the workload. That is where Chinese price pressure becomes dangerous.

DeepSeek V4-Pro also arrives as xAI launches Grok 4.6 and as Anthropic’s investors reportedly talk about extraordinary IPO valuations. The contrast is striking. Western AI headlines are about trillion-dollar valuations and massive compute plans. Chinese AI headlines are increasingly about capability per dollar.

There are caveats. OpenRouter pricing may differ from direct provider pricing, benchmark results can vary and developers will need real-world testing before switching serious workloads. But the signal is still clear. DeepSeek wants to make powerful AI cheaper, and that forces everyone else to explain why their premium pricing is worth it.

The next stage of the AI race may therefore be less glamorous than model demos. It may be a price war over inference. If DeepSeek and other Chinese labs keep driving costs down while improving capability, the biggest winners may be developers and businesses. The biggest losers may be frontier labs whose valuations assume customers will keep paying premium prices forever.