Alibaba.com Says Accio Cuts Commerce Agent Costs by More Than 50%

Alibaba.com said in a press release that Accio completed a benchmark covering 107 online commerce tasks at more than 50% lower estimated cost than OpenAI Codex and Anthropic Claude Code, with comparable completion quality. Alibaba.com designed and ran the benchmark, and Accio is its own product. Completing all tasks cost an estimated $3.69 with Accio, compared with $9.27 for Codex and $9.51 for Claude Code.

Alibaba.com attributed the difference to task specific model and resource selection, cache reuse, context compression, and coordination between agents. The company said no single model led across every task in the benchmark. The announcement was made at CoCreate, Alibaba.com's own annual event for entrepreneurs and small businesses.

Alibaba.com open sourced Commerce Agent Bench on GitHub, so the tasks and scoring can be inspected and rerun by others. The benchmark draws on activity from 10 million active small and medium businesses, 1.6 million conversations, and 200,000 execution traces.

The company also expanded the Accio workspace with tools for market research, product selection, product development, supplier evaluation, and daily operations. It connects with supported storefront workflows on Amazon, Shopify, eBay, TikTok Shop, and Walmart.