Every CaptchAPI figure below was measured on the production VPS that serves your traffic, over the public tunnel, with warm instances. Competitor figures are the averages those providers publish for themselves. We have not re-run their solvers, and we say so on every row.
We are fastest on four of the six. On Turnstile proxyless and reCAPTCHA v2 we are not, and the tables show it.
The most commoditised scenario, and the one where the field is tightest. AnyFlare and CapSolver are around a second ahead of us. A second matters less here than it looks — all three are under four seconds — but we are not going to call third place a win.
| Provider | Median solve time | Published success rate |
|---|---|---|
| AnyFlareprovider's published figure | 2.7s | 96.7% |
| CapSolverprovider's published figure | 2.8s | 98.3% |
| CaptchAPImeasured by us · 5/5 | 3.6s | — |
Routing a real browser through your proxy is where most providers lose time, because they re-launch a context per request. Our warm pool applies the proxy per context instead, which is where the 2.5s gap comes from.
| Provider | Median solve time | Published success rate |
|---|---|---|
| CaptchAPImeasured by us | 6.9s | — |
| CapMonsterprovider's published figure | 9.4s | 96.2% |
| AnyFlareprovider's published figure | 9.6s | 85.1% |
The widest margin we measure, and the number the landing page leads with. v3 is score-based with no interaction, so it rewards a warm real browser and punishes providers that farm it out to workers.
| Provider | Median solve time | Published success rate |
|---|---|---|
| CaptchAPImeasured by us · 4/4 | 1.84s | — |
| SolveCaptchaprovider's published figure | 6.2s | — |
| CapSolverprovider's published figure | 6.8s | — |
| Castlebreakerprovider's published figure | 10.2s | — |
We solve v2 through the audio challenge, which is bounded by how long Google takes to serve and validate the clip. CapMonster is faster here. If v2 is the bulk of your volume, that is a fair reason to test both.
| Provider | Median solve time | Published success rate |
|---|---|---|
| CapMonsterprovider's published figure | 7.9s | — |
| CapSolverprovider's published figure | 9.0s | — |
| CaptchAPImeasured by us | 9.5s | — |
The "Just a moment…" interstitial, returning cf_clearance cookies and the exact User-Agent that earned them. Roughly half the time of the published alternatives, at a higher success rate than either.
| Provider | Median solve time | Published success rate |
|---|---|---|
| CaptchAPImeasured by us · 3/3 | 4.8s | — |
| CapMonsterprovider's published figure | 9.3s | 91% |
| EZCaptchaprovider's published figure | 10.0s | 94.7% |
The spread here is enormous because success rate and speed trade off hard. Our 6.2s is the median; the tail runs to 33s on multi-round challenges. Even the tail beats AntiCaptcha's published average.
| Provider | Median solve time | Published success rate |
|---|---|---|
| CaptchAPImeasured by us | 6.2s | — |
| VoidSolverprovider's published figure | 12.7s | 66% |
| Viperprovider's published figure | 16.9s | 24% |
| AntiCaptchaprovider's published figure | 50.2s | 90% |
A benchmark page that wins every row is a marketing page. These are the cases where the measurement did not go our way, stated plainly so you can weigh them against your own mix.
AnyFlare (2.7s) and CapSolver (2.8s) are ahead of our 3.6s. If your workload is almost entirely proxyless Turnstile and you are optimising the last second, they are a reasonable choice. Add a proxy requirement and the ranking inverts.
CapMonster's published 7.9s beats our measured 9.5s. The audio challenge has a floor we have not broken through yet.
AnyFlare publishes roughly 1.5s against our measured 4.8s on the JSD sensor variant specifically. Our full-page challenge number is the strong one; JSD is not.
Asking for a real Gecko engine costs us about 5 extra seconds on Turnstile (9.7s), because launching Camoufox is not free. It is a genuine Firefox token rather than a spoofed header, which is the trade we chose.
Five dollars of credit is enough to reproduce every row on this page against your own targets. That is a better benchmark than ours.