Service status

Status

Computed from settled traffic, not from a probe we run against ourselves. Windows with no traffic report no reading rather than a perfect score.

Go to the market
Success rate, 24h
88.40%
707 settled turns
Success rate, 72h
87.94%
5315 settled turns
Failed turns
641
Over the last 72h
Online lanes
20866
Across 384 models
Last 72 hours, hour by hour
ServedFailedNo traffic
72h agoNow
By model, last 7 days
ModelTurnsFailedSuccess rate
gpt-6-luna3,83517995.33%
gpt-6-astra1,09513487.76%
gpt-5.6-terra1,0358292.08%
claude-sonnet-58221897.81%
gpt-5.6-luna6273195.06%
claude-haiku-4-55751896.87%
gpt-6.1-sol5514292.38%
gpt-5.55331697.00%
claude-opus-5-54931996.15%
claude-opus-5447997.99%
claude-sonnet-5-5403798.26%
claude-fable-53933591.09%
claude-fable-5-13543490.40%
gpt-6-sol3261994.17%
kimi-k2.7-code2201792.27%
claude-haiku-5-5182796.15%
kimi-k2.6180298.89%
kimi-k3169298.82%
seedance-2.5980100.00%
kimi-k2-thinking86297.67%

23 more models with traffic are not listed.

Incidents

Nothing recorded. When something goes wrong — a degradation, an outage, a billing error — it gets written here with what happened and why, not only fixed quietly.

How to read this
  • A turn counts once it has settled — done or failed. Requests still in flight are excluded, so the right edge of the chart is not dragged down by traffic that has not finished yet.
  • Nothing here is synthetic. Every figure comes from traffic real buyers paid for; we run no self-probe, because a platform that grades its own test always passes.
  • A window with no traffic shows no reading, never 100%. Quiet and healthy are different facts.
  • This measures the relay: whether a request reached a seller's upstream account and came back. It does not measure the upstream vendors' own availability, and it is not a judgement of any individual seller.

Last computed Oct 11, 2026, 1:05 AM.