ZoomInfo Launches GTM Bench, First AI Benchmark Designed for Real-World Sales and Marketing Work

ZoomInfo Introduces GTM Bench to Evaluate AI on Go-to-Market Performance

ZoomInfo has released GTM Bench, an AI benchmark specifically designed to evaluate large language models and AI agents on real-world GTM tasks that sales and marketing teams perform daily. Version 1 of this benchmark for AI sales tools covers more than 20 jobs, 4 systems, and 3 models, with the methodology, sample tasks, and grading rubrics published for public scrutiny2

2

. The benchmark tests AI systems on building target lists, enriching records, scoring accounts, and reaching decision-makers—the core activities that drive revenue for B2B organizations.

How GTM Bench Differs from Traditional AI Benchmarks

Most AI benchmarks measure reasoning within closed environments, providing models with all necessary facts upfront. But go-to-market work fails differently. The primary constraint is data availability: roughly 70% of B2B contact data decays annually, and critical information sits scattered across the web and private systems. While a summary that's 90% accurate might still prove useful, a prospect list with the same accuracy rate sends sales representatives to wrong companies. This reality shaped how ZoomInfo designed its AI-powered go-to-market platform evaluation system.

Every GTM Bench result is graded against a senior GTM operator's work product on two independent axes. Answer measures what share of the requested work product the system delivers, while grounding measures what share of returned data traces to a real, current source. A confident wrong answer scores negative, ensuring systems are penalized for hallucinations or outdated information.

ZoomInfo's GTM.AI Dominates Initial Testing Results

In the version 1 run, ZoomInfo's GTM.AI led every pillar with a GTM Bench Index of 77, compared to 47 for Apollo, 36 for Exa, and 31 for open-web search. The system completed 98% of the operator's work product and returned 478 verifiable records per 1,000, vastly outperforming competitors who returned between 7 to 35 verifiable records. On the same 1,000-contact suite, non-ZoomInfo systems returned 720 wrong phone numbers. The cost efficiency also favored ZoomInfo at $0.79 per task2

2

.

Transparency and Limitations Acknowledged

ZoomInfo acknowledges GTM Bench is a vendor-run benchmark and publishes results accordingly. The company openly displays losses, including four categories where its advantage is thin or absent—from pure copywriting to owned CRM data that no external tool can access. Grounding is graded against ZoomInfo's verified records, representing a reach measure rather than an independent audit2

2

. The benchmark is versioned and will be re-run on major model releases, with version 2 adding agentic multi-step workflows, international coverage, and an owned-data axis.

What Powers GTM.AI and Its Growing Ecosystem

GTM.AI serves as ZoomInfo's headless GTM context layer, exposing the GTM Context Graph—which includes 100 million companies, 500 million contacts, and billions of signals—through API and Model Context Protocol. Every record carries a confidence score and its lineage, addressing the verifiable data challenge central to revenue-driving tasks. The system powers dozens of live integrations, including Salesforce Agentforce, HubSpot Breeze, Microsoft Copilot, Anthropic's Claude, ChatGPT, Gong, LeanData, and Google Workspace. Other data providers and agent builders can submit their systems for evaluation. ZoomInfo will run them at their best configuration and add results to the leaderboard2

2

. The full task set, the 1,000-contact grounding study, and the 20-job tool-choice study are available on request. This open approach aims to establish GTM Bench as an industry standard for evaluating AI performance on actual go-to-market work, helping sales and marketing teams make informed decisions about which AI tools deliver reliable results rather than impressive demos.