Multilingual local assistants
Build text assistants that can run without sending every request to a cloud model.
Independent tool overview
Tiny Aya at a glance
Tiny Aya is a compact Cohere Labs model family built for translation and text generation across 70 languages on local, mobile, edge, and hosted environments.
Visit the official Tiny Aya site ↗
- Model size
- 3.35B parameters
- Languages
- 70
- Context window
- 8K tokens
- Variants
- Base, Global, Earth, Fire, and Water
- Access
- Cohere API and Hugging Face weights
- Weight license
- CC-BY-NC 4.0 plus acceptable-use policy
Overview
What Tiny Aya is
Tiny Aya is a family of 3.35-billion-parameter multilingual language models from Cohere and Cohere Labs. It supports 70 languages in a much smaller footprint than typical frontier models, with an emphasis on translation, multilingual understanding, target-language generation, and practical local deployment.
The family includes a pretrained base model, a globally balanced instruction model, and three region-specialized instruction variants. Earth focuses on West Asian and African languages, Fire on South Asian languages, and Water on European and Asia-Pacific languages. All variants use an 8K context window.
Developers can download the open weights and GGUF quantizations from Hugging Face or call the instruction variants through Cohere's Chat API. The weights are licensed under CC-BY-NC 4.0 plus Cohere Labs' acceptable-use policy, so they are suitable for research and non-commercial projects—not unrestricted commercial deployment.
Use cases
Who Tiny Aya is best for
The strongest fit depends on the job you need the product to complete, not the size of its feature list.
Translation and localization
Generate and translate content across 70 supported languages, including many lower-resource languages.
Research and adaptation
Study multilingual representations or adapt the base and instruction weights for non-commercial downstream work.
Mobile and edge prototypes
Use compact or quantized builds where privacy, offline use, latency, or limited compute matters.
Capabilities
Core Tiny Aya features
Broad multilingual coverage
Supports 70 languages spanning Europe, Africa, West Asia, South Asia, Southeast Asia, and East Asia.
Region-specialized variants
Offers Earth, Fire, and Water checkpoints tuned for stronger performance in different language regions.
Globally balanced model
Tiny Aya Global is designed to provide the best overall balance across languages and regions.
Local open weights
Provides base and instruction-tuned checkpoints through Hugging Face for local inference and research.
GGUF quantizations
Official GGUF variants make it easier to run the instruction models with efficient local inference tools.
Hosted Chat API
All four instruction-tuned variants can be called through Cohere's Chat endpoint and SDKs.
Process
How the Tiny Aya workflow works
Step 1
Choose the deployment path
Use Cohere's API for a hosted experience or download the weights when local processing and model control matter.
Step 2
Select a variant
Start with Global for broad coverage or choose Earth, Fire, or Water when the target languages align with a regional specialization.
Step 3
Check the license
Confirm that a non-commercial CC-BY-NC release fits the intended use before downloading or adapting the weights.
Step 4
Pick model precision
Use the full checkpoint for maximum fidelity or a GGUF quantization when device memory and speed are tighter.
Step 5
Evaluate every target language
Test fluency, translation fidelity, hallucinations, cultural context, and safety separately for the languages that matter.
Step 6
Measure the actual device
Benchmark memory use, latency, battery impact, and output quality on the intended phone, laptop, or edge hardware.
Cost
Tiny Aya pricing and free plan
Tiny Aya's weights can be downloaded without a model fee for non-commercial use under CC-BY-NC 4.0, though local compute and hosting still cost money. Cohere trial API keys are free but limited. Cohere does not publish a separate production token price for Tiny Aya on its public pricing page, so production users should check the dashboard or contact Cohere.
Open-weight download
No model fee for permitted non-commercial use
Download Tiny Aya checkpoints and GGUF files from Hugging Face under the published license and acceptable-use policy.
- CC-BY-NC 4.0
- Commercial use is not granted
- Local hardware costs apply
- Base and instruction variants available
Cohere trial API
Free with limits
Evaluate instruction-tuned Tiny Aya models through Cohere's Chat API using a trial key.
- Rate limited
- Not permitted for production or commercial use
- Chat endpoint
- Account required
Cohere production API
Check current terms
Production access is metered, but a Tiny Aya-specific public token rate is not listed on Cohere's pricing page.
- Production key required
- Pay-as-you-go billing
- Confirm model availability and rates
- Contact Cohere for production needs
Pricing checked . Check current pricing at the source ↗
Assessment
Tiny Aya strengths and limitations
Where it stands out
- Unusually broad language coverage for a 3.35B-parameter model
- Regional variants address the uneven quality common in multilingual models
- Small enough for realistic local and edge experimentation
- Hosted API, full weights, and GGUF files provide several deployment paths
- Research release supports adaptation and multilingual open-science work
What to consider
- The CC-BY-NC license does not permit unrestricted commercial use
- An 8K context window is short compared with many current cloud models
- Fluent output can still contain fabricated or incorrect facts
- Quality and safety can vary substantially by language, dialect, and code-mixed prompt
- Running on a phone depends on quantization, memory, runtime support, and the specific device
Compare
Tiny Aya alternatives
The right alternative depends on the specific output, workflow, controls and budget your project requires.
Consumer
Qwen3.5 Small
Consider Qwen3.5 Small for a different family of compact open-weight models with broad general capabilities.
Explore Qwen3.5 Small →Coding
Meta Llama Models
Consider Meta Llama for a larger ecosystem of local models, runtimes, fine-tuning tools, and deployment providers.
Explore Meta Llama Models →Coding
Ollama
Consider Ollama as a convenient local runtime for supported open-weight models when ease of setup matters most.
Explore Ollama →Questions
Tiny Aya FAQs
What is Tiny Aya?
Tiny Aya is a Cohere Labs family of 3.35B-parameter text models designed for multilingual understanding, translation, and generation across 70 languages.
Is Tiny Aya open source?
It is more precise to call Tiny Aya open-weight. The weights are available under CC-BY-NC 4.0 and Cohere Labs' acceptable-use policy, which restrict commercial use.
Which Tiny Aya model should I use?
Use Global for the broadest regional balance, Earth for West Asian and African languages, Fire for South Asian languages, and Water for European and Asia-Pacific languages. Base is intended for pretraining-oriented research and adaptation.
Can Tiny Aya run on a phone?
Cohere designed it for local and on-device deployment and provides GGUF quantizations. Real-world performance depends on the device, quantization, inference runtime, and requested context length.
Can I use Tiny Aya through an API?
Yes. Tiny Aya Global, Earth, Fire, and Water are available through Cohere's Chat API.
How much does Tiny Aya cost?
The weights have no model download fee for permitted non-commercial use, while local compute is your responsibility. Trial API access is free but limited; Cohere does not list a separate public production token rate specifically for Tiny Aya.
Does Tiny Aya support images?
The official Tiny Aya family described here is text-only. Tiny Aya Vision is a separate community research project and should not be assumed to be part of the standard released checkpoints.
Bottom line
Our Tiny Aya verdict
Tiny Aya is a strong research and prototyping choice when broad multilingual coverage must fit into a small local model. The biggest caveats are the non-commercial license, short 8K context, and the need to evaluate accuracy and safety independently in every target language.
Visit Tiny Aya website ↗Similar Consumer Tools
View all Consumer AI tools →
Claude Opus 4.6
Claude Opus 4.6 - Anthropic's upgrade to its most powerful model line, featuring multi-agent collaboration, 1M context window, and new Office integrations
Qwen3-TTS CustomVoice 1.7B
Qwen3-TTS CustomVoice 1.7B - Alibaba's multilingual text-to-speech model with voice cloning using just three seconds of reference audio
Step-3.5-Flash
Step-3.5-Flash - StepFun's powerful open-source model with strong reasoning and agentic capabilities
Claude Sonnet 4.6
Claude Sonnet 4.6 - Anthropic's upgraded mid-tier model rivaling Opus at 1/5 the cost with 1M-token context
AI training for the future of work.
Get access to all our AI courses, hundreds of real-world AI use cases, live expert-led workshops, an exclusive network of AI early adopters, and more.
AI Courses
Get unlimited access to all of our current & upcoming industry-specific AI courses for the duration of your subscription.
Daily Guides
To keep up with the rapid pace of AI, our team publishes AI implementation guides daily. Our library contains 300+ practical use cases to automate real-world work.
Workshops
Join weekly, live, interactive sessions with industry leaders who are at the forefront of AI for hands-on implementation guidance and exclusive insights.
Community
Network with an exclusive community of AI-first professionals who are working smarter with AI. Learn how early adopters are using AI in their work and businesses.