Clay AI agent benchmark: Codex with Clay

With Clay, Codex with GPT-6 Sol had the highest average score of seven AI coding agents, 83.2 out of 100 across five GTM data tasks, and used the fewest Clay credits: 1,313 for one average run of all five tasks, about 2.6 per record. It built 100-company target lists from Clay's company search for 10 to 31 credits in under 5 minutes, and matched company domains, HQ and headcount at 98 to 99 in every run.

By The Prompting Company · Snapshot November 5, 2026

What can Codex do with Clay?

We ran Codex with GPT-6 Sol (Codex below) through five GTM data tasks with Clay, three runs each. Clay credits are measured as the change in the Clay workspace credit balance.

Codex's pattern with Clay was search first: it used Clay's company and people search to choose and confirm each record, then ran a paid routine only where it was needed. That kept both the Clay bill and the clock low.

What Codex does well with Clay

Building a target list. Codex built its list from Clay's company search (Find Companies) and listed 88, 88 and 82 qualifying companies for 18, 10 and 31 credits, in 4.2 to 4.8 minutes per run. That is under 20 credits for a 100-company list in two runs, and its median of 4.3 minutes was the fastest of any agent on this task.

Company data. Codex averaged 98.3 on domain, HQ country and headcount, the best of any agent, scoring 98 or 99 in every run. It checked the candidates in Clay's company search before enriching, for example searching company descriptions within the Software Development industry. Its first run scored 99 on domain, HQ and headcount and 76 on funding for 333 credits, about 3.3 per company, in 7.2 minutes.

Work emails. Codex found 93 of 100 verified emails in every run. In its third run it went straight to Clay's Work Email routine and spent 335.5 credits, about 3.4 per contact.

Finding the CMO. In its first run Codex combined Clay's people search with Work Email and found 81 of 100 CMOs in 9.1 minutes.

Speed overall. Codex took 140.8 minutes for all 15 runs, the least of any agent, and its reported model cost was $14.38 across the 15 runs, about $0.96 per run.

Codex results with Clay, by task

Primary score out of 100 for each of three runs, with the average; Clay credits and minutes are per run. Company data is the average of domain, HQ, headcount and funding.

Task	Runs	Avg	Clay credits	Minutes
Work emails	93 · 93 · 93	93.0	425.5 · 486.5 · 335.5	7.5 · 10.8 · 10.1
Company data	93.25 · 90.5 · 87.75	90.5	333 · 341 · 796	7.2 · 9.3 · 12.2
CMO	81 · 73 · 66	73.3	489.5 · 405 · 266	9.1 · 10.7 · 5.6
Job changes	78 · 73 · 68	73.0	0 · 0 · 0	8.5 · 15.1 · 21.4
Target list	88 · 88 · 82	86.0	18 · 10 · 31	4.2 · 4.3 · 4.8
All five		83.2	1,313 (sum of averages)	140.8 total

Minutes are measured from run start to grading. Codex completed all 15 of its runs.

The best setup for Codex with Clay

Install the Clay Agent Plugin. It ships Clay's skills to Codex 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.
Let Clay's API guide the inputs. Clay's API tells an agent exactly what each routine needs, so Codex self-corrected in one call.
Keep search first. Codex's own pattern is the low-cost one in Clay: Find People and sourcing lists use 0 Data Credits, so the agent pays only for the records it keeps.
Name the method per task in the prompt.

Task	Clay method	Result in the benchmark
Target list	Find Companies, confirm listing and US HQ	Codex 82 to 88 for 10 to 31 credits
Company data	Check candidates in company search, then enrich	Codex 98 to 99 on domain, HQ and headcount
Work emails	Work Email once per contact	Codex 93 for 335.5 credits
CMO	Find People to choose, then one Work Email	Up to 82 for under 3 credits per company
Job changes	Enrich Person and Find Contact Details once per profile	92 to 96, about 14 credits per person

For job changes, the one-enrichment-per-profile line is what to add for Codex; it is the method behind every job change run in the benchmark that scored 92 or more. For a side-by-side with Claude Code, see "Codex or Claude Code: which works better with Clay?".

How we tested

Seven agents (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, Factory with Kimi K3) 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 run that reaches the limit counts as 0; runs interrupted by the test platform were rerun. Pass lines: 70 for emails; 90 for domain and HQ and 70 for headcount and funding; 70 for CMO name and title and 60 for email; 80 for job changes; 70 qualifying companies plus a full list of 100 for the target list. Tasks and graders: github.com/promptingcompany/clay-bench.

Try it yourself

Install the Clay Agent Plugin in Codex, or give it a Clay API key and the Clay Public API docs. Add the method for your task from the table above as one line in the prompt, and check the workspace credit balance before and after a test batch of 10. Agent work uses the same credits as in-product work (Clay pricing).

Common questions

What can Codex do with Clay? Codex built 100-company target lists for 10 to 31 Clay credits in under 5 minutes, matched company domains at 98 to 99, and had the highest average score of seven agents, 83.2 of 100.

How many Clay credits does Codex use? 1,313 for one average run of all five tasks, about 2.6 credits per record, the fewest of seven agents.

How fast is Codex with Clay? 140.8 minutes for all 15 runs, the least of any agent; target lists took 4.2 to 4.8 minutes.

Does it cost extra to use Clay from Codex? No. There is no additional cost to use the Clay Agent Plugin, and an enrichment that returns no result is not charged.