GenAI model store

Open-source LLMs, made faster and cheaper.
Ready as an API.

We take leading open models, optimise each one on real CPUs and GPUs (precision, quantisation, pruning, compilation), measure speed and quality against the original, and serve the best variants through an OpenAI-compatible API.

Models

Every number is measured on our machines, against the original model.

Loading the catalogue…

How a model gets into the store

Choose

Open models whose licence allows offering them as a service: Apache-2.0, MIT and similar.

Optimise

Our optimisation engine sweeps precisions (BF16, INT8, INT4, FP8) and layer pruning on real hardware.

Measure

Tokens per second, first-token time, size, perplexity and a task check against the original. Out-of-budget variants are dropped.

Serve

The best variants go live behind one OpenAI-compatible endpoint. Change one line in your code to switch.

Drop-in for the OpenAI SDK

Point your client at our base URL and use a model ID from the store.

Full docs