How we counted the benchmark results
By solvethree Updated
The 96.0% on our homepage comes from 1,440 accepted requests out of 1,500 attempts. We ran 500 request-only solves for each of three workflows in September 2026. A pass meant the real downstream target accepted the generated token.
On this page
What counted as a pass
We tested an airline loyalty login, live ticket inventory, and membership signup. Each workflow had 500 attempts, and the target's response determined the outcome.
Getting a token back from a solver wasn't enough. The request had to reach the downstream target and have that token accepted. That's the distinction behind the word acceptance on the homepage.
The results, workflow by workflow
The denominator includes all 1,500 attempts: 1,440 accepted, 19 challenged and 41 rejected.
| Workflow | Accepted | Challenged | Rejected | Acceptance |
|---|---|---|---|---|
| Airline loyalty login | 489 | 7 | 4 | 97.8% |
| Live ticket inventory | 500 | 0 | 0 | 100.0% |
| Membership signup | 451 | 12 | 37 | 90.2% |
| All three workflows | 1,440 | 19 | 41 | 96.0% |
Where the 96.0% comes from
Each workflow contributes the same 500 attempts, so the overall result isn't weighted toward the easiest target. You can also see the variation directly: the membership signup result was 90.2%, while ticket inventory reached 100.0% in this sample.
(489 + 500 + 451) / (500 + 500 + 500) = 1,440 / 1,500 = 96.0%The comparison on the homepage
The homepage comparison uses the same targets and 500 solves per site. It reports 96.0% for solvethree, 48.2% for CapSolver, and 24.5% for 2Captcha in the September 2026 sample.
These figures describe those workflows at that time. Before moving a large workload, try the service on the site and action you actually need to solve.
Why we don't name the targets
Target identities are withheld for responsible disclosure. The workflow labels describe what was tested without publishing the sites or their protected endpoints.