Aya Vision 8B
Aya Vision 8B is the smaller research checkpoint for multilingual image and text understanding. It covers 23 languages and remains available as weights after the hosted 8B API retired.
Research release
Capabilities and limits
Its architecture combines a language model derived from Command R7B with a SigLIP2 vision encoder and a multimodal adapter. The model card covers OCR, captioning and visual reasoning.aya-vision-8b official model card
The Cohere c4ai-aya-vision-8b API retired on 4 April 2026. Downloadable weights have their own license and access terms.Cohere model and API deprecations
When this model fits
Practical guidance. Use this checkpoint for local experiments where multilingual visual understanding is the subject of the work. Review tiny text, crowded images and unfamiliar layouts rather than relying on simple photo captions. Compare with North Micro Vision Instruct if permissive weight licensing is important. Compare with the hosted 32B model if the priority is API access rather than operating a local checkpoint.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | c4ai-aya-vision-8b |
| Model scale | 8 billion parameter classaya-vision-8b official model card |
| Context | 16K tokensaya-vision-8b official model card |
| Weight license | CC-BY-NC 4.0aya-vision-8b official model card |
Related models
- Aya Vision 32B
Cohere · Aya
A 32-billion-parameter vision-language model with image and text input and text output in 23 languages.
Launch date not confirmedAPI · Downloadable A 2.4-billion-parameter vision-language model for OCR, charts and document questions, with interleaved image and text input.
Launch date not confirmedDownloadable