Aya Vision
Open-weight multilingual vision-language models (8B/32B) for research use.
About
Cohere's nonprofit research lab released Aya Vision in 8B and 32B sizes to close the multilingual gap in vision-language models: 23 languages, built on Aya Expanse/Command R backbones with a SigLIP 2 vision encoder, trained with synthetic annotations and cross-modal model merging.
The important catch is the license: CC-BY-NC-4.0 plus an acceptable-use addendum, so this is research-only, not commercial open source. Benchmark wins are Cohere-reported on Cohere's own AyaVisionBench, which warrants the usual skepticism.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- 8B beats Qwen2.5-VL-7B on AyaVisionBench (Cohere-reported; benchmark developed by Cohere)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (verified tier, metrics-v1). Evidence: Cohere for AI release, 2025-03-04; 8B and 32B variants; CC-BY-NC-4.0 + acceptable-use addendum — research only, NOT commercial open source; Benchmark wins are Cohere-reported on Cohere's own AyaVisionBench. License: CC-BY-NC-4.0 confirmed from multiple sources — non-commercial use only..
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 7.0/10
- Maintenance · recency, releases, issue load
- 7.0/10
- Openness · license + docs
- 5.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · Cohere for AI release, 2025-03-04; 8B and 32B variants
- · CC-BY-NC-4.0 + acceptable-use addendum — research only, NOT commercial open source
- · Benchmark wins are Cohere-reported on Cohere's own AyaVisionBench