Try Delta
← All models

Tiny Aya

Cohere · 2025 · Compact multilingual · regional variants

Cohere’s compact multilingual family. A 3.35B-class base with Global, Earth, Fire, and Water regional variants. 70+ languages without a cloud round-trip.

Maker
Cohere
Architecture
Compact multilingual · regional variants
Fits from
Phone / edge · 6 GB+ RAM
Context
up to 131K
Runs
Local Intelligence
MultilingualEdge70+ languagesRegional variantsAny device

Sizes

Tiny Aya GlobalQ4_K_M

requires 6 GB+ memory · 131K context · Phone / edge

2.1 GBWeights
Tiny Aya EarthQ4_K_M

requires 6 GB+ memory · 131K context · Phone / edge

2.1 GBWeights
Tiny Aya FireQ4_K_M

requires 6 GB+ memory · 131K context · Phone / edge

2.1 GBWeights
Tiny Aya WaterQ4_K_M

requires 6 GB+ memory · 131K context · Phone / edge

2.1 GBWeights

Memory at context length

Approximate working set for Tiny Aya Global as context grows.

4K
2.7 GB
8K
2.8 GB
16K
3 GB
33K
3.2 GB
66K
3.3 GB
131K
3.5 GB

Related families

Ministral 3

Mistral’s edge family with native vision: phones through workstations, fully local.

Gemma 3N

Google’s efficient open model, tuned to run locally on phones, tablets, and laptops.

Gemma 4

Google’s open models for phones and IoT: audio, vision, and near-zero latency on your silicon.