DeepSeek Sets Cost Record as Chinese Open Models Close AI Gap

Chinese open-source AI models are rapidly narrowing the performance gap with frontier proprietary systems from US developers including OpenAI and Anthropic. The shift matters for companies weighing model vendors and deployment strategies, particularly as agentic AI moves beyond simple prompts to multi-step workflows. In that setting, benchmark scores alone offer an incomplete picture: total task cost and response speed increasingly determine whether a model is commercially useful at scale.

As of Aug. 12, 2026, the latest Artificial Analysis evaluation showed DeepSeek-V4 Flash delivering performance comparable with GPT-5.6 Luna while costing just $0.03 per task, the lowest among models in its class. The result gave DeepSeek the strongest price-performance showing in the comparison and underscored how Chinese developers are competing on economics as well as capability, potentially strengthening their appeal for enterprise agentic workflows.