Models

Every size in the lineup, grouped by tier. They all share one architecture and differ only in depth and width.

Tier
Status for
No.ModelParametersLayers × widthContextStatus
Tiny50M – 125M · Phones, browsers, embedded devices
01WrenOur smallest model, built for the browser and tiny devices.50M20 × 3842,048Planned v1Planned v1
02FinchA step up from Wren, with more room for vocabulary and facts.85M21 × 5122,048Planned v1Planned v1
03RobinThe largest of the tiny tier and a good default for on-device tasks.125M27 × 5762,048Planned v1Planned v1
Small160M – 350M · Phones and laptops
04LarkEntry to the small tier, with a longer 4K context window.160M29 × 6404,096Planned v1Planned v1
05KestrelA balanced, general-purpose model that runs well on a phone.250M26 × 8964,096Planned v1Planned v1
06OspreyThe strongest small model, for when quality matters more than speed.350M31 × 9604,096Planned v1Planned v1
Medium500M – 900M · Laptops and single GPUs
07HeronHalf a billion parameters, comfortable on any modern laptop.500M31 × 1,1524,096Planned v1Planned v1
08CraneStronger reasoning and writing, still on a single consumer GPU.700M33 × 1,3444,096Planned v1Planned v1
09CondorOur largest model and the flagship of the lineup.900M33 × 1,5364,096Planned v1Planned v1
PlannedIn trainingReleased