Models
Every size in the lineup, grouped by tier. They all share one architecture and differ only in depth and width.
| No. | Model | Parameters | Layers × width | Context | Status |
|---|---|---|---|---|---|
| Tiny50M – 125M · Phones, browsers, embedded devices | |||||
| 01 | WrenOur smallest model, built for the browser and tiny devices. | 50M | 20 × 384 | 2,048 | Planned v1Planned v1 |
| 02 | FinchA step up from Wren, with more room for vocabulary and facts. | 85M | 21 × 512 | 2,048 | Planned v1Planned v1 |
| 03 | RobinThe largest of the tiny tier and a good default for on-device tasks. | 125M | 27 × 576 | 2,048 | Planned v1Planned v1 |
| Small160M – 350M · Phones and laptops | |||||
| 04 | LarkEntry to the small tier, with a longer 4K context window. | 160M | 29 × 640 | 4,096 | Planned v1Planned v1 |
| 05 | KestrelA balanced, general-purpose model that runs well on a phone. | 250M | 26 × 896 | 4,096 | Planned v1Planned v1 |
| 06 | OspreyThe strongest small model, for when quality matters more than speed. | 350M | 31 × 960 | 4,096 | Planned v1Planned v1 |
| Medium500M – 900M · Laptops and single GPUs | |||||
| 07 | HeronHalf a billion parameters, comfortable on any modern laptop. | 500M | 31 × 1,152 | 4,096 | Planned v1Planned v1 |
| 08 | CraneStronger reasoning and writing, still on a single consumer GPU. | 700M | 33 × 1,344 | 4,096 | Planned v1Planned v1 |
| 09 | CondorOur largest model and the flagship of the lineup. | 900M | 33 × 1,536 | 4,096 | Planned v1Planned v1 |
PlannedIn trainingReleased