← All leaderboards
Screenshots
What this board tests: rendered image of a url. Every provider below ran the identical corpus on 2026-09-14. The composite blends measured quality, latency p50, the track-specific cost metric and error rate. Method and corpora: how we measure.
| Rank | Provider | Composite | Valid image | Cost | Latency p50 |
|---|---|---|---|---|---|
| 1 | 9/10 | 1.00 | $0.0032 / call | 5.1 s | |
| 2 | 8.92/10 | 1.00 | $0.002 / call | 6.4 s | |
| 3 | 8.52/10 | 0.88 | $0.00098 / call | 4.2 s | |
| 4 | 7.95/10 | 0.88 | $0.0009 / call | 13.1 s | |
| 5 | 7.64/10 | 0.75 | $0.001 / call | 6.1 s | |
| 6 | 7.51/10 | 0.75 | $0.01 / call | 5.2 s | |
| 7 | 7.09/10 | 0.75 | $0.0049 / call | 34.8 s | |
| 8 | 2.56/10 | 0.12 | $0.0014 / call | 2.2 s |
Low scores are published alongside high ones — a ranking you can only trust if nothing gets hidden. No per-call markup, no commission from any provider. Cost figures are priced by the 2026-09-14 run's own table; the price the gateway bills today is on each provider page.