Clay AI agent benchmark: Clay credits per work email

With Clay, an AI agent found verified work emails for 93 of 100 contacts for 337.5 Clay credits: about 3.4 credits per contact searched, or 3.6 per email found. In a benchmark of seven AI coding agents, three runs each, the cheapest runs called Clay's Work Email routine about once per contact and skipped the company lookups. Running extra lookups added credits without finding a single extra email.

By The Prompting Company · Snapshot November 5, 2026 · Three runs per agent

What does one work email cost in Clay?

We gave seven agents the same 100 contacts, each with a first name, last name and company domain, and asked them to fill in work emails using only Clay. Six of the seven found the same 93 verified addresses in every run. Every agent stayed under 5 credits per correct email, and the lowest-cost runs came from the simplest workflow.

Clay credits are measured as the change in the Clay workspace credit balance. Pi spent exactly 337.5 credits in all three of its runs.

What did the lowest-cost agents do?

Pi with GPT-5.6 Luna and Grok CLI with Grok 4.6 used a two-step workflow:

1. Run Clay's Work Email routine once for every contact, passing the name, the domain, and the domain again as the company name.
2. Write what Clay returned, and leave the cell empty when it returned nothing.

The routine asks for a company name, which the list didn't have. Passing the domain in that field worked: one Claude Code with Sonnet 5 run tested both ways and found it did not change Clay's answer.

Clay's Work Email routine is a waterfall across data providers, and Clay states that "you only pay credits for the provider that finds a match" (see Clay waterfall enrichment). Clay's pricing page adds that an enrichment with no result is not charged.

What keeps the cost at about 3.4 credits per contact?

Sending each contact straight to Work Email. The routine does not need a company lookup first, and Clay does not charge when it returns nothing, so there is no need to retry blanks. Codex with GPT-6 Sol went straight to Work Email in its third run and spent 335.5 credits for 93 emails; Factory with Kimi K3's first run spent 334.5. Optional steps, such as looking up every company with Enrich Company first or re-running the contacts Clay left empty, add credits without adding emails.

The same routine also handles a missing company name. Passing the domain in that field, as described above, let six of the seven agents find the same 93 emails in every run. More detail is in "How accurate is Clay for AI agents finding work emails?"

Does the model bill matter too?

Yes, and here the agent with the lowest average Clay bill was also the cheapest model to run. Pi's model cost was about 1 cent per run, against 22 to 88 cents for the other agents, for the same 93 emails and the lowest average Clay credits.

Batch size made no difference to the Clay bill: ten batches of 10, two of 50, or a test of 5 then 95 all cost 337.5. That is covered in "Does batching Clay API calls change the credit cost?"

How we tested

Seven agents ran the task three times each: Claude Code with Opus 5.5 and Sonnet 5, Codex with GPT-6 Sol, Pi with GPT-5.6 Luna, Grok CLI with Grok 4.6, OpenCode with Kimi K3 and Factory with Kimi K3. Each run was alone in a fresh sandbox with the same Clay key and the same 19 saved Clay routines, and a 60-minute limit; a timeout would count as 0, and runs stopped by a platform failure were rerun. All 21 runs finished. Agents ran one at a time, so each run's Clay credits are the exact change in the workspace balance. Every address in the answer key was verified deliverable, and an email counts only when it matches exactly. A model judge confirmed every value came from Clay and nobody was emailed.

This is one list of 100 contacts across Australia, Canada, the US and Europe. A list with different countries or cleaner names would land differently.

Try it yourself

The task and grader are at github.com/promptingcompany/clay-bench. To find work emails for your own list, install the Clay Agent Plugin or give your agent a Clay API key and the Clay Public API docs, and tell it: run the Work Email routine once per contact, pass the domain as the company name if you don't have one, and don't look up companies or retry blanks first. The Clay Agent Plugin ships Clay's skills to your agent up front, including the habits that keep credit use down, and setup is one prompt (about 15 minutes). Clay states there is no additional cost to use the Plugin; agent work uses the same credits as work in Clay itself.

Common questions

How many Clay credits does it cost an AI agent to find a work email? About 3.4 credits per contact searched, or 3.6 per email found: 337.5 credits for 100 contacts and 93 verified emails in every Pi run. The cheapest single runs, Factory Kimi's first and Codex's third, spent 334.5 and 335.5.

Does a company lookup first help? No. The Work Email routine does not need one: Codex spent 335.5 credits for 93 emails when it went straight to Work Email. Extra lookups add credits without improving results.

Does Clay charge for contacts it can't find? Clay's pricing page states that an enrichment with no result is not charged, and its Work Email waterfall charges only for the provider that finds a match.

Which agent was cheapest overall? Pi with GPT-5.6 Luna: 337.5 Clay credits in every run and about 1 cent of model cost.