Alibaba's Accio Open-Sources CommerceAgentBench for E-commerce Agents

Alibaba's Accio team open-sourced CommerceAgentBench, a benchmark of 107 real e-commerce tasks run on stateful replicas of live services. It evaluates agent actions rather than outputs, with Claude completing just over half the tasks and Qwen topping open-source models.

2026-08-30 ~ 2026-09-01 · 3 related posts

• Accio Open-Sources CommerceAgentBench, Claude Passes 52% — dr_cintas · 2026-08-30
• CommerceAgentBench Open Source: Evaluating Agent Actions, Not Outputs — SucceededMind · 2026-08-31
• CommerceAgentBench released: Qwen leads open-weight models — Alibaba_Qwen · 2026-09-01