Overview
Chinese AI developers have strengthened their position in the emerging market for agentic artificial intelligence, capturing three of the top five global spots for agentic performance, according to a research note from Jefferies.
Jefferies findings and model rankings
Analyst Edison Lee summarized the trend succinctly, saying that "China continues to catch up." Jefferies recorded that six of the ten best-performing models worldwide are now Chinese, up from five in a prior assessment, and highlighted that among the five models judged to have the best agentic capabilities, three are Chinese.
The report also quantified the relative closeness of top-tier models: "the agentic performance gaps between Chinese and US models in the top 5 are small, ranging from 2% to 12%." That narrow spread, Jefferies said, positions Chinese firms to benefit as adoption of agentic systems increases.
Adoption and enterprise signals
Jefferies pointed to surveys conducted by Google Cloud and Anthropic showing "rising enterprise usage (50%+) of agents." The firm noted that agentic AI - which moves beyond answering questions to performing tasks - has accelerated quickly and dominated recent industry discussions because of its potential to automate personal and business workflows.
However, the note emphasized that enterprise adoption is not yet straightforward. As Lee put it, "ROI is still hard to quantify, and multiple challenges exist." Jefferies concluded that while agents have transformed AI capabilities, "widespread enterprise deployment will take time due to reliability issues and uncertain inference costs."
Commercial dynamics and cost
Price was identified as a competitive lever for Chinese systems. Jefferies said token costs for Chinese models are a fraction of those charged by U.S. models, a cost differential that is helping Chinese offerings expand their global download share. The note singled out Alibaba's Qwen as the leading model, with its international adoption "steadily rising."
Market context and investor tools mentioned
The note also included investor-oriented material raising the question "Should you be buying 9988 right now?" and described an AI-driven stock screener called ProPicks AI that evaluates companies, including 9988, across many financial metrics. The promotional item cited past winners identified by that tool.
Key points
- Chinese models now occupy three of the top five slots for agentic AI performance, and six of the top ten overall - indicating a narrowing performance gap with U.S. models.
- Surveys show enterprise usage of agents exceeding 50%, but ROI remains difficult to quantify; this affects enterprise software and cloud service procurement decisions.
- Lower token costs for Chinese models are expanding their global download share, bolstering international adoption for offerings such as Alibaba's Qwen.
Risks and uncertainties
- Return on investment for agentic AI is still hard to quantify - introducing uncertainty for enterprise buyers and vendor revenue models.
- Reliability issues could slow broad enterprise deployment, affecting enterprise software adoption rates and service-level expectations.
- Uncertain inference costs make total cost of ownership unclear for large-scale deployments, impacting cloud economics and budgeting.