Clay AI agent benchmark: what Pi can do with Clay

With Clay, Pi with GPT-5.6 Luna found 93 of 100 verified work emails in every run for 337.5 Clay credits, about 3.4 per contact, and about 1 cent of model cost per run. That matched the top work email score in our benchmark at the lowest model cost of seven agents. Pi was also the fastest agent overall, and it checked 100 contacts for job changes in 4.7 minutes at about 14 credits per person.

By The Prompting Company · Snapshot November 5, 2026

What does Pi do well with Clay?

We ran seven AI coding agents through five GTM data tasks with Clay, three runs each. Pi pairs a very small model bill with short, direct runs. Clay credits are measured as the change in the Clay workspace credit balance.

Work emails at the top score for about a cent. Pi sent each contact to Clay's Work Email routine once and found 93 of 100 verified emails in all three runs, the top score on the task. It spent exactly 337.5 credits each time, the lowest average of any agent, and its model cost was $0.01 per run. With Clay's Work Email waterfall, you only pay credits for the provider that finds a match.

The lowest model cost in the benchmark. Pi's model cost for all 15 runs was $0.25, from $0.01 to $0.03 per run. For comparison, 86 of the other six agents' 87 reported runs each cost more than $0.25 on their own.

Fast runs. Pi had the lowest sum of median run times of any agent, 34.5 minutes for one typical run of each of the five tasks. It had the lowest median of all seven agents on company data, 4.8 minutes, and on job changes, 4.9 minutes. It ran Clay's Enrich Person and Find Contact Details on all 100 LinkedIn profiles in under 5 minutes in two job change runs, and its 85-of-100 run took 4.7 minutes.

The same result run after run. Pi repeated its work email score and its Clay spend exactly in all three runs. On company data, its scores of 83 on domain, 90 on HQ country and 91 on headcount held in every run, and each run took 3.8 to 6.9 minutes.

CMOs for under 3 credits per company. In its third CMO run, Pi used Clay's people search to choose each person, then ran Work Email for that person: 69 of 100 CMOs named for 288 credits, in 7.4 minutes.

The best setup for Pi with Clay

Pi works best when the prompt names the Clay method. These are the methods behind the top runs in the benchmark:

1. Work emails: run Work Email once per contact, and pass the domain as the company name if you have no company name. This is Pi's own method.
2. Job changes: run Enrich Person and Find Contact Details once per LinkedIn URL, then write a stayed or left decision for every row. Runs that did this and decided nearly every row scored 92 to 96.
3. CMOs: choose the person with Find People, which uses 0 Data Credits, then run one Work Email lookup for that person.
4. Company data: confirm each company in Clay's company search before enriching, and take funding rounds from Clay's company news.
5. Target lists: start from Find Companies with the "Public Company" company type, then confirm each company is listed in its own name and headquartered in the US before adding it.

A rule of one paid routine per record keeps Pi's Clay bill as low as its model bill; running extra lookups added credits without improving results.

Pi's results across the five tasks

Primary score out of 100: emails correct, the average of domain, HQ, headcount and funding, CMO name correct, people correctly marked as stayed or left, and qualifying companies on the target list.

Model cost per run is the average of three runs. For how Pi compares on cost and speed, see "What is the cheapest AI agent to run with Clay?" and "Which AI agent finishes Clay tasks fastest?".

How we tested

Seven agents ran each of five tasks three times, one agent at a time, in a fresh sandbox with the same Clay key, the same 19 saved Clay routines and a 60-minute limit. A timeout counts as 0, and platform failures were rerun. Each task had 100 records. Minutes run from start to grading, and model cost is as reported by the test platform. Tasks, prompts and graders are at github.com/promptingcompany/clay-bench.

Try it yourself

The Clay Agent Plugin ships Clay's skills to your agent up front, including the habits that keep credit use down; setup is one prompt (about 15 minutes), and there is no additional cost to use it. Install the Clay Agent Plugin in Pi, or give Pi a Clay API key and the Clay Public API docs. Add the five method lines above to your prompt and run a test batch of 10 first. Agent work uses the same credits as in-product work (Clay pricing).

Common questions

What can Pi do with Clay? Pi with GPT-5.6 Luna found 93 of 100 verified work emails with Clay in every run for 337.5 credits and about 1 cent of model cost, and it was the fastest of seven agents across five tasks.

How much does it cost to run Pi with Clay? Pi's model cost was $0.25 for all 15 runs. On the Clay side, its runs cost about 3.4 credits per contact for work emails and about 14 credits per person for job changes.

How fast is Pi with Clay? Its median runs of all five tasks added up to 34.5 minutes, the lowest of seven agents, and it checked 100 LinkedIn profiles for job changes in 4.7 minutes.

Which Clay tasks suit Pi best? Work emails and job change checks, where one Clay routine per record does the work. For Pi against Codex, see "GPT-6 Sol or GPT-5.6 Luna: which OpenAI model works better with Clay?".