Serve any model behind one API.
140+ open and proprietary models, routed per request for quality, latency and cost. Swap models with a string.
Four things you stop building yourself.
Routing, streaming, caching and regions ship with the API. There is no extra service to run.
Smart routing
Set a quality threshold once. Lattis picks the model per request and falls back automatically.
Streaming everywhere
Text, audio and vision streams through one interface and one client.
Caching
Semantic and exact caches cut repeat cost by up to 70%.
Regions
11 regions, with data residency pinned per API key.
Switch models with a string, not a migration.
Point the client at "auto" and set the bar. Pin a model when you need to, and change it later without touching the rest of your code.
Measured, not promised.
p50 across 1M production requests, September 2026. Methodology in the docs.
“We did not add capacity. We added Lattis.”
Which models are available?
140+ open and proprietary models across text, code, vision and audio. The catalogue updates every week.
Can I bring my own fine-tuned model?
Yes. Connect a fine-tune from a supported provider or upload your own weights. It gets the same endpoint, routing and traces as every other model.
How does routing decide?
Each request goes to the cheapest model that clears your quality threshold within your latency budget. Every decision is written to the trace.
What happens when a provider has an outage?
Requests fail over to an equivalent model in under 200 ms, with no code changes. The trace shows which model answered.
Start with 1M free tokens.
No card required. Production keys in two minutes.