Chinese AI models are rapidly narrowing the performance gap with high-performance frontier models from developers like OpenAI and Anthropic, while delivering overwhelming cost efficiency measured by cost per task. According to an analysis by Artificial Analysis (AA), an independent AI evaluation site, DeepSeek V4 Flash offered the lowest cost per task among the top 20 models as of Aug. 7, at just 3 cents. Most Chinese AI models are released as open-weight models. While they previously competed primarily on low pricing, recently released models such as Moonshot AI's Kimi K3 and Alibaba's Qwen 3.8 Max have now entered the top 10 in AA's composite Intelligence Index. Expanding the list to the top 20 also includes DeepSeek V4 Flash, released late last month. Notably, these Chinese models qualify as upper-tier frontier or near-frontier systems under the AA Intelligence Index while maintaining exceptionally low costs per task. As surging AI transformation (AX) spending becomes a major enterprise concern, AA's latest cost-per-task metric is drawing significant attention. According to Forbes in May, Uber exhausted its entire 2026 annual AI budget by April after developer usage of Anthropic's Claude Code surged. As Claude Code spread across roughly 5,000 engineers, AI expenses far exceeded original expectations. Uber subsequently introduced a monthly spending cap of $1,500 per employee for agentic coding tools. Tech giant Amazon faced a similar issue. According to TechRadar earlier this month