Chinese AI model DeepSeek-R1 matches U.S. rivals on benchmark tests at lower cost

A Chinese AI model called DeepSeek-R1 has posted benchmark scores that match or approach those of models from Anthropic and OpenAI while offering substantially lower usage fees. The model is available through an application programming interface that charges a fraction of the rates set by the two U.S. companies.

Developers in the United States have begun testing the system on standard evaluation suites used to compare large language models.

Performance and pricing On several widely used benchmarks, DeepSeek-R1 produced results within a few percentage points of the top U.S. models, according to test data shared by users on public forums. The pricing difference remains the clearest distinction: the Chinese service lists rates that are reported to be less than one-tenth of those charged by the leading American providers.

Researchers who have run the model locally on consumer-grade hardware note that it requires less memory than comparable systems, though exact hardware specifications vary by deployment.

Market response U.S.-based developers have posted code samples and performance logs showing the model handling tasks such as code generation and document summarization at speeds similar to domestic alternatives. Some users have begun routing portions of their workloads to the lower-cost service while retaining U.S. models for other tasks.

No official statements from Anthropic or OpenAI regarding the new competitor have been issued.