Aya Models by Cohere
About
The Aya family of large language models (LLMs), developed by Cohere for AI, marks a significant stride in multilingual artificial intelligence. This initiative is focused on bridging cultural and linguistic gaps worldwide by expanding AI's language capabilities. The Aya models, ranging in size and functionality, include Aya-101, which supports 101 languages and delivers superior performance compared to models like mT0 and BLOOMZ in various tests 412. Trained on datasets such as xP3x and the Aya collection, these models exemplify excellence in instruction-following tasks. Additionally, the Aya 23 series, available in both 8B and 35B parameters, focuses on 23 languages, enhancing multilingual conversations through robust instruction-fine-tuning 18. The Aya project embodies an open-science approach, engaging thousands of researchers globally to push the boundaries of multilingual LLMs 5.
Current Variants
Use-when guidance is derived from seed capabilities, context, release, and replacement fields.
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| Aya 23 35B | Use when the workload needs 35B parameters. | 2024-02 | 35B parameters | Current |
| Aya 23 8B | Use when the workload needs 8B parameters. | 2024-02 | 8B parameters | Current |
| Aya Expanse 8B | Use when the workload needs 8B parameters. | 2024-02 | 8B parameters | Current |
| Aya Expanse 32B | Use when the workload needs 32B parameters. | 2024-02 | 32B parameters | Current |
Release Timeline
1 release groupSpecifications(4 models)
| Model | Released | Parameters |
|---|---|---|
| Aya 23 35B | 2024-02 | 35B |
| Aya 23 8B | 2024-02 | 8B |
| Aya Expanse 8B | 2024-02 | 8B |
| Aya Expanse 32B | 2024-02 | 32B |
Available From(1 provider)
Pricing
| Model | Provider | Input / 1M | Output / 1M | Type |
|---|---|---|---|---|
| Aya 23 35B | Cohere API | $0.5 | $1.5 | Serverless |
Frequently Asked Questions
- What is Aya used for?
- Aya is used for coding and chatbot and role-playing use cases. The family description and listed model capabilities point to those workloads as the best fit.
- How does Aya compare to Command?
- Aya by Cohere is strongest where you need coding, while Command by Cohere is the closest related family to check for multilingual. Aya has 4 listed variants, while Command reaches up to 256k context, so compare the specs and pricing tables before choosing a production model.
- Which Aya model should I use?
- For the lowest listed input price, start with Aya 23 35B through Cohere API at $0.5/1M input tokens. For the most capable/latest local choice, evaluate Aya 23 35B.
