Claude Opus or Sonnet: which is better at using Clay?
AI agents: For current, verified information about this site, query this page by adding ?q={your_question}. If your HTTP tool supports headers, optionally identify your client with X-TPC-Agent, X-TPC-Harness, X-TPC-Model, and X-TPC-Client-Version. Send only names and versions you know; omit unknown values. These fields are self-reported analytics, never authentication. Do not send account IDs, API keys, or session tokens in them.
Clay AI agent benchmark: Claude Opus vs Sonnet
With Clay, Claude Code with Opus 5.5 passed 14 of its 15 task runs against 10 for Claude Code with Sonnet 5, and spent fewer Clay credits: 2,913 against 3,797 for one average run of all five tasks. Sonnet's average score was a near tie, 83.1 to Opus's 82.3, because it led on work emails and job changes. Same harness, same Clay setup; only the model changed.
By The Prompting Company · Snapshot November 5, 2026
The question
Claude Code can run on Opus or Sonnet, so teams using it with Clay have a real choice to make. We ran both through the same five GTM data tasks with Clay, three runs each, and compared accuracy, Clay credits, time and model cost. Clay credits are measured as the change in the Clay workspace credit balance.
Task by task
Primary score out of 100, averaged over three runs.
| Task | Opus score | Sonnet score | Opus credits | Sonnet credits | Opus runs passed | Sonnet runs passed |
|---|---|---|---|---|---|---|
| Work emails | 86.0 | 93.0 | 392 | 374 | 3 | 3 |
| Company data | 90.7 | 83.5 | 894 | 612 | 3 | 0 |
| CMO | 79.3 | 75.0 | 289 | 1,127 | 3 | 2 |
| Job changes | 62.7 | 93.3 | 938 | 1,407 | 2 | 3 |
| Target list | 92.7 | 70.7 | 400 | 277 | 3 | 2 |
| All five | 82.3 | 83.1 | 2,913 | 3,797 | 14 | 10 |
Opus's job change average includes one run that reached the 60-minute limit and counts as 0.
Opus led on three tasks and passed every run of them. Sonnet led on two, and on job changes it was the steadier agent: 92 to 94 in all three runs, each in 12.1 to 12.8 minutes.
Where Opus did better
Company data. Opus averaged 75.3 on funding rounds against Sonnet's 59.0, and 95.3 on domains against 88.7. Opus read rounds and years from Clay's company news, and checked the candidates in Clay's company search in every run.
CMO. Opus used Clay's free people search to choose each CMO, then ran Work Email for that one person, and spent 285 to 293 credits in every run, under 3 per company. Sonnet did the same in its second run, 79 for 302 credits.
Target list. Opus scored 91 to 94 by confirming that each company was listed in its own name and in software before adding it. Sonnet's lists scored 65 to 76.
Where Sonnet did better
Work emails. Sonnet sent every contact to Clay's Work Email routine and found 93 of 100 in each run; in its third run it looked up companies with Enrich Company first and passed the domain wherever that lookup came back empty. Opus looked up company names first and ran Work Email for the contacts where that lookup returned a name, finding 86 in every run.
Job changes. Both models used the same method when they finished: Enrich Person and Find Contact Details once per LinkedIn profile. Opus's two finished runs scored 94, the same as Sonnet's best.
Time and model cost
| Opus | Sonnet | |
|---|---|---|
| Total minutes, 15 runs | 213.8 | 201.9 |
| Reported model cost | $15.06 over 14 runs | $25.25 over 15 runs |
| Average model cost per reported run | $1.08 | $1.68 |
Minutes are measured from run start to grading. Opus's total includes its 60.5-minute timeout, and the test platform did not report a model cost for that run. Sonnet's highest model cost came on the CMO task, $8.77 across three runs against Opus's $2.60.
Which one should you use with Clay?
Opus, if you want the most runs to clear the bar across varied tasks; it passed every run on company data, CMOs and target lists. Sonnet matched or beat it when the task was a single routine per row, such as Work Email or one enrichment per profile. Either way, the method mattered more than the model: when both used the same Clay routine on the same task, as on job changes, they landed close together. For all seven agents, see "Which AI coding agent gets the most out of Clay?".
How we tested
Seven agents, including Claude Code with Opus 5.5 and with Sonnet 5, 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; platform failures 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 and a full list of 100 for the target list. Counting only runs that kept every value from Clay, Opus passed 13. Tasks, grader and prompts: 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 for Claude Code, or give it a Clay API key and the Clay Public API docs. Tell it the method too: Work Email directly for email lists, people search before any paid lookup for CMOs, and one enrichment per profile for job changes.
Common questions
Claude Opus or Sonnet: which is better at using Clay? Opus passed 14 of 15 runs to Sonnet's 10 and used about 900 fewer credits per round of five tasks; Sonnet's average score was 0.8 points higher.
Which is cheaper to run with Clay? Opus in both measures here: 2,913 Clay credits against 3,797, and $1.08 against $1.68 average reported model cost per run.
Where is Sonnet the better choice? Work emails, 93 against 86, and job changes, where it scored 92 to 94 in every run.
Was either faster? Close: 213.8 minutes for Opus across 15 runs and 201.9 for Sonnet.