Latest Data from Alibaba International Digital Commerce Group's Accio: AI Cost for Full E-Commerce Task Execution Is Over 50% Lower Than General-Purpose Solutions

Recently, at CoCreate, Alibaba International Digital Commerce Group's annual event for entrepreneurs and small businesses, the company released benchmark test results for 107 tasks on its commercial AI agent platform Accio. According to the data, Accio's estimated total cost to complete the full suite of e-commerce tasks is $3.69, compared to $9.27 for OpenAI Codex and $9.51 for Anthropic Claude Code over the same period. Accio's costs are more than 50% lower than both, while delivering comparable task completion quality, making it the most cost-competitive e-commerce AI tool for small and medium-sized merchants currently available.

Zhang Kuo, President of Alibaba International Digital Commerce Group, noted that AI that is unaffordable for small businesses is completely useless. The team's goal is not simply to improve AI performance, but to make commercial AI sufficiently practical and affordable even for one-person companies.

As introduced, Accio's efficiency stems from full-system optimization tailored to real business scenarios. The platform leverages e-commerce-specific data and workflows to conduct targeted post-training on lightweight models, adapting them to clearly segmented tasks. Small models can handle routine work efficiently, while more capable models are called upon only when deep reasoning is required. Accio does not simply select the cheapest model, nor does it default to calling the most powerful model. Instead, it breaks down complex demands into executable steps, balancing quality, speed, cost and data requirements for each step, so that merchants do not have to pay for unneeded computing power resources. Mechanisms such as cache reuse, context compression and collaborative agent execution also reduce duplicate processing, unnecessary tool calls and redundant token consumption.

At the event, Alibaba International Digital Commerce Group also launched an upgraded version of Accio, which integrates research, sourcing and daily e-commerce operation functions to form a unified AI workspace for small businesses. Merchants can complete market research, product opportunity identification, product development, supplier evaluation and daily operation management through Accio. The platform has been integrated with channels including Amazon, Shopify, eBay, TikTok Shop and Walmart, allowing merchants to complete store operations supported by the above platforms on a single interface.

In addition, Alibaba International Digital Commerce Group has open-sourced the Commerce Agent Bench test suite on GitHub. Instead of using synthetic training questions, the test suite extracts 107 end-to-end e-commerce tasks covering 7 categories and 4 autonomy levels from the real behavior of merchants, including 10 million active small and medium-sized merchants, 1.6 million conversations and 200,000 execution traces. Specific tasks include reviewing hundreds of unstructured emails, identifying payment fraud, calculating landed costs, and booking multi-carrier shipping routes, among others. The test results show that no single model can maintain a leading position across all task categories, further validating the rationality of task-level routing – that is, different tasks are best handled by different AI models, eliminating the need to rely on a single general-purpose model to complete all work.

This article was first published on the official Ebrun website.