Visa launched a la benchmark for “agentic commerce” that tested 22 AI shopping agents across 15 tasks to evaluate their ability to purchase products based on specific user requirements. The la report found that these AI agents succeeded in just 57.2% of the test runs on average, meaning nearly half of the purchase attempts failed.
Among the tools tested, Google’s Gemini 4 Argon and Muse Wonder recorded the highest accuracy with an 82% success rate. The la results also detailed the operational costs for the top performers, noting that Gemini 4 Argon cost $0.54 per run while Muse Wonder cost $1.55.
