LLM cost & latency benchmarks

Real, anonymized numbers aggregated across every request processed through CostTrim in the last 30 days — no workspace, user, or prompt content included, just cost, latency, and error rate by provider, model, and use case. Server-rendered and refreshed on every visit.

Is a provider slow right now?

Latency and error rate over the last 24 hours, refreshed on every page load.

Not enough requests in the last 24 hours yet (0 of 20 needed) to show an honest live snapshot — see the 30-day view below instead.

Monitor latency and error rate for your own app — free — with CostTrim.

Not enough traffic yet to publish an honest benchmark (0 of 100 requests needed in the last 30 days) — small samples would just show one workspace's test traffic, not a real pattern. Check back soon, or start sending traffic through CostTrim yourself.