Chinese Models Close the Gap: The Premium Paid for Frontier Models Comes Under Question

The most striking development in the AI market over the past two months has been the speed and price of models coming out of China. They are closer to the frontier than ever, which is leading people to question the high prices and valuations of companies like OpenAI and Anthropic.

The gap in numbers

• Chinese models trail the best US models by 1% to 16% on current leaderboards, roughly a 7–8 month gap in time.
• Open Chinese models can be 60–90% cheaper than leading Anthropic and OpenAI models.
• In about two months, five companies shipped six near-frontier models: Kimi K3, Qwen3.8-Max, DeepSeek V4-Flash and V4 Pro, GLM-5.2 and ByteDance's video model Seedance.
• In some areas, such as coding and agent benchmarks, Chinese models have drawn level with US models.

The DeepSeek example: cheap isn't the whole story

DeepSeek's annualized revenue run rate rose from $500 million to $1 billion in a few months, almost entirely from API usage. What stands out is that demand stayed strong even though the company raised API prices 2.3x to 4.5x. It is now preparing to close a funding round of about $7.5 billion at a $75 billion valuation.

The real question for enterprises

A cheaper model is tempting, but price can't be the only criterion. Which data goes to a provider in which country, the model's license, its security evaluation and regulatory obligations such as KVKK or GDPR all have to be weighed. A practical approach: send non-sensitive, high-volume work to cheap models, and route work involving personal data or trade secrets to approved providers or to open models running on your own infrastructure.

The remot angle

remot makes it possible to make this choice for each request: depending on the data's sensitivity, region and cost target, a request goes either to a cheap open model or to an approved frontier model. That way you capture the savings of cheaper models without losing control over which data goes where.