Benchmarks

Benchmarked on our hardware. Not datasheets.

Single-stream generation speed from real workloads on real machines — the worst case, not batched multi-tenant numbers. Every result below was measured on a dedicated server physically hosted in Egypt.

1× RTX 4090

Up to 102 tok/s · 128K ctxUpdated 2026-07-19 · Full results

2× RTX 4090

Up to 183 tok/s · 128K ctxUpdated 2026-07-19 · Full results

Why single-stream numbers?

Most hosting benchmarks quote batched throughput — many requests added together to make one big number. Your users don't arrive in batches. We publish the speed a single request gets on an otherwise idle machine, because that's the experience you can actually expect.