ZoomInfo has launched GTM Bench, a versioned benchmark designed to evaluate large language models (LLMs) and AI agents on real-world go-to-market (GTM) activities rather than conventional reasoning tasks. Version 1 assesses performance across more than 20 GTM jobs, four systems, and three models, with publicly available methodology, sample tasks, and grading rubrics to promote transparency. Unlike traditional AI benchmarks that test reasoning using predefined facts, GTM Bench focuses on the accuracy and reliability of live business data, addressing challenges such as the rapid decay of B2B contact information and fragmented data sources. Results are measured on two key metrics-Answer, which evaluates how much of the requested task is completed, and Grounding, which measures whether returned data is accurate, current, and verifiable.
Also Read: IBM Advances Enterprise AI Software to Multi-Agent Workflows and AI Development
In the inaugural benchmark, ZoomInfo’s GTM.AI achieved the highest GTM Bench Index score of 77, outperforming Apollo (47), Exa (36), and open-web search (31). It completed 98% of the evaluated work, delivered 478 verifiable records per 1,000, and achieved the lowest cost per task at $0.79, while competing systems returned significantly more inaccurate contact data. ZoomInfo also disclosed benchmark limitations, including categories where its advantage was minimal, and outlined plans for Version 2, which will introduce agentic multi-step workflows, international coverage, and an owned-data evaluation axis. The benchmark is powered by GTM.AI, ZoomInfo’s AI-driven GTM context layer, which leverages its GTM Context Graph and integrates with enterprise platforms such as Salesforce Agentforce, HubSpot Breeze, Microsoft Copilot, Claude, ChatGPT, Gong, LeanData, and Google Workspace.