Alibaba's E-Commerce AI Agent Costs 61% Less Than Anthropic's, Attracting Over 60,000 Paying Users in Five Months
nashnova research
Alibaba.com's AI agent product Accio has signed over 60,000 paying users in five months, putting its ARR above $60 million. A new open-source benchmark shows Accio completes the same 107 e-commerce tasks at 39% of Anthropic's token cost — this means the e-commerce AI race is shifting from "can it work" to "can you afford to run it."
What has Accio achieved in five months?
Accio launched in April and has attracted over 60,000 paying users, each spending an average of more than $1,000.
That translates to annualized recurring revenue (ARR) — projected full-year subscription income at the current run rate — of over $60 million.
This means → a B2B platform's bolt-on AI product has reached mid-size SaaS revenue scale in five months, validating real willingness to pay for AI in e-commerce.
61% cheaper than Anthropic — how was it measured?
Alibaba.com president Zhang Kuo unveiled Commerce Agent Bench, a benchmark covering 107 common e-commerce tasks — supplier sourcing, complex shipping-route planning, and more.
Token cost to complete all tasks: Accio $3.69, OpenAI Codex $9.27, Anthropic Claude Code $9.51.
In plain terms = for the same workload, Accio spends less than 40% of what Anthropic's solution costs; that gap compounds fast when an agent runs thousands of tasks a day.
Zhang emphasized the benchmark is fully open-sourced — anyone can verify the results.
Why can it run so much cheaper?
Accio automatically routes each user to the most cost-effective AI model and has fine-tuned selected models on e-commerce tasks — retraining general-purpose models with industry data so they perform better on trade-specific jobs.
Zhang's logic: e-commerce is a high-competition, margin-compressing industry where AI affordability is critical. "Your agent might complete a task flawlessly in a lab, but if it needs to do that thousands of times a day, cost must be controllable."
This reflects a deeper bet: in high-frequency, low-margin scenarios like e-commerce, cost itself is product competitiveness, not a nice-to-have.
Who is the direct rival, and where is the battlefield?
Accio's main competitor is Anthropic's Claude Commerce Agent, launched in early September, targeting both buyers and merchants with e-commerce agent capabilities.
The two will clash head-on in the U.S. market — this year's CoCreate conference drew over 10,000 American buyers, nearly triple last year's attendance.
Alibaba.com has more than 8 million users in the U.S.; its parent global wholesale segment posted $2.1 billion in revenue last quarter, up 7% year-on-year.
What comes next to prove?
The cost advantage is on the table. The core question now: can Accio build a differentiated moat from Alibaba's proprietary industry data, accumulated since 1999?
In plain terms = being cheap is the entry ticket; the real barrier is e-commerce domain knowledge that general-purpose AI labs cannot build or buy — supplier matching, cross-border logistics planning, trade compliance.
This means → the e-commerce AI agent race will shift from "whose model is smarter" to "who understands the industry better" — exactly the card Alibaba wants to play.
市场有风险，内容仅供研究参考，不构成投资建议。