ZoomInfo Technologies Inc. has announced the launch of its GTM Bench benchmarking system, designed to evaluate the performance of large language models and artificial intelligence agents in real-world go-to-market tasks.

The GTM Bench scores AI systems across two primary dimensions: "Task Completion," which measures the proportion of requested work the system accomplishes, and "Data Provenance," which measures the proportion of returned data that is traceable to real, current sources. In the inaugural round of testing, ZoomInfo's own GTM.AI achieved a composite index score of 77, compared to Apollo's 47, Exa's 36, and open web search's 31. The ZoomInfo system completed 98% of operational tasks and could return 478 verifiable records per thousand, significantly outperforming other systems which returned between 7 and 35. Its cost per task was just $0.79.

It is worth noting that ZoomInfo also transparently disclosed four categories where its own performance was subpar, including areas like pure copywriting and accessing proprietary CRM data that external tools cannot reach. The measurement of data provenance is also based on ZoomInfo's own verification records, rather than being conducted through an independent audit.

A second version of GTM Bench will introduce new evaluation dimensions, including multi-step agent workflows, international coverage, and proprietary data assessment. Other data providers and AI agent developers will also be able to submit their systems for testing. GTM.AI serves as ZoomInfo's contextual data layer, providing access to over 100 million companies, 500 million contacts, and tens of billions of signal data points. It is already integrated with platforms including Salesforce Agentforce, HubSpot Breeze, Microsoft Copilot, Claude, and ChatGPT.