İvmeLabs builds AI models that stay small but do one thing — or a few, if we're feeling ambitious — right. İvme covers our production model families, each with a fruit codename; Expİvme is where we test architectures and training styles that haven't earned a name yet.
Prediction and conversation using human language. The flagship line, currently on v3.
The first series built around a single category — writing code at incredibly small sizes.
Routes prompts to the correct model in a given set, using their metrics.
Zero-shot classification at small sizes.
Polishes user text on the go: fixing grammar, adding markdown styling.
Autocorrect and word prediction, specifically for Turkish.
TTS models with realistic-sounding voices at small parameter sizes, for assistants you hear constantly.
Our first MoE series. The first model lands at 1.2B — not really small, but with a quirk we're not ready to talk about yet.
İvmeLabs