Chinese AI models from developers such as DeepSeek and Z.ai are capturing significant market share from OpenAI and Anthropic by offering substantially lower costs while maintaining competitive performance. Data from OpenRouter shows Chinese models now account for more than 30% of weekly enterprise token processing in the US, up from an average of 11% over the prior year, with some open-source Chinese models priced 60-90% cheaper than leading US models. Startups such as Lindy have shifted their traffic entirely to cheaper Chinese alternatives, while Z.ai’s GLM 5.2 has drawn attention for matching Anthropic’s Opus on some benchmarks at roughly a fifth of the cost.
