SteadyGatewayRequest access

Steady under load

Method first. Numbers when they are real.

A performance claim without a published method is marketing. We commit to the method before any number appears, so that when results land you know exactly what they mean.

Five measurements, published in full.

  1. Latency

    Time-to-first-token at the 50th, 95th and 99th percentile from real probe locations in Europe, the US and India — segmented by model, region and request type, compared against direct provider access.

    Reporting soon

  2. Throughput

    Sustained capacity under reservation, behaviour at the ceiling, and the error curve under concurrency. Published as full curves rather than a single headline number.

    Reporting soon

  3. Availability

    A public status record on a rolling 90-day window. Contractual availability commitments reference that record — we make no public availability claim until it exists.

    Reporting soon

  4. Output consistency

    A fixed prompt set run on schedule against pinned versions, with published scoring — evidence that what you pinned is what you keep getting.

    Reporting soon

  5. Version freshness

    How long each provider takes to make a newly released model available, sourced and dated per entry. Live now.

    Open the availability ledger →

    Live

Rules we can be held to.

Every number is dated and sourced. Anything older than 90 days is shown as stale — it degrades by build rule, not by our diligence on a good day.

Percentiles, not averages. Enterprise workloads are hurt at the 99th percentile, so that is the number we lead with.

No selective windows. Published series run continuously. A bad week stays in the record.

Ask for the current benchmark scope.

Pilot customers measure with their own traffic — the only benchmark that predicts your workload.

Talk to our teamhello@steadygateway.com

99.9% availability commitment with tiered service credits · Reply within one business day · NDA available on request