Chinese competition and open-source models are driving down AI API costs, but the decision goes beyond token pricing

The cost of hiring artificial intelligence through APIs is rapidly decreasing as new players enter the market. Chinese-developed models, such as those from DeepSeek, and open-source solutions are now directly competing with companies that traditionally charged higher prices for access to their technologies.

For contractors, the message is clear: comparing only the price per token is no longer sufficient to determine which solution is worthwhile.

Why prices are falling

The competition gained momentum when Chinese companies began offering models capable of performing tasks similar to proprietary alternatives at a fraction of the cost. For instance, Moonshot AI’s Kimi K3 model delivers performance comparable to a top-tier closed model but at a price that is approximately 30% lower.

A September report by Mozilla highlights that open-source models can be downloaded and adapted, although this does not mean that all the code, training data, and development processes are available.

The cost goes beyond the token

For a company, the actual cost depends on how AI is used. For example, an application that processes millions of documents can generate a very different expense compared to a chatbot used occasionally by employees.

The competition itself is shifting the focus of comparisons. An analysis by Bank of America shows that the Chinese market is beginning to value the total cost of completing a task more than the isolated price of each token.

More advanced models continue to charge more when they offer complex reasoning, extensive context, multimodal resources, or the ability to execute workflows with multiple agents.

This dynamic creates opportunities for a hybrid architecture. Repetitive and less complex tasks can be assigned to cheaper alternatives, while jobs requiring greater capacity remain in premium solutions.

What changes for AI buyers

Competitive pressure increases companies’ choices. It is now possible to evaluate different options based on cost, performance, security, support, and infrastructure needs.

However, there is a difference between accessing a model through a platform and operating it internally. Open-source models can offer greater control, but organizations still need to consider hardware, technical staff, maintenance, and security.

Mozilla also notes that closed models remain attractive to organizations that require dedicated support, regulatory compliance, and a ready-to-use solution.