Aya Vision 32B on live video

What Aya Vision 32B is good at

Aya Vision 32B was trained explicitly for multilingual coverage across 23 languages, which shows up as consistent captioning and visual question answering quality regardless of which language a user asks in. That makes it a natural fit for global products where users interacting with a live camera or shared screen aren’t all writing in English.

Beyond raw language coverage, Aya Vision 32B handles cross-lingual visual QA well, meaning the image content can be described in one language while the question and answer happen in another, without a separate translation step.

• Multilingual captioning and visual question answering across 23 languages
• Cross-lingual QA where the query language differs from the response language
• General-purpose visual understanding for global, non-English-first products

Aya Vision 32B and the Overshoot API

Aya Vision 32B is not currently in Overshoot's live model catalog; the authoritative list is GET /v1beta/models. If a multilingual model like this were available through the API, the workflow would be the standard one: publish a camera or screen share over WebRTC to open a Stream, then send a chat-completions request whose image_url is an ovs:// reference to the latest frame or a recent segment. Multilingual training like Aya Vision 32B's would let one live stream serve users asking questions in different languages without switching models.

For models that do run on Overshoot's hosted infrastructure, answers stream back over SSE with low time to first token, and reusing thread_id across a session keeps prompt caching effective for repeated queries against one stream.

Aya Vision 32B next to Command A Vision

Cohere’s vision lineup also includes Command A Vision, a larger 112B model tuned for enterprise document and chart work with a commercial API alongside its open weights. Aya Vision 32B is the smaller, more research-oriented sibling, prioritizing broad multilingual coverage over the enterprise document depth that Command A Vision targets.

Frequently asked questions

Can Aya Vision 32B analyze live video in multiple languages?

Aya Vision 32B is built for multilingual visual question answering across 23 languages, but it is not currently in Overshoot's live model catalog. When a model like this is available through the API, you reference a live WebRTC stream with an ovs:// URL in a chat-completions request and the answer streams back over SSE; check GET /v1beta/models for what is live.

Is Aya Vision 32B free to use commercially?

Aya Vision 32B is released under CC-BY-NC 4.0, a non-commercial research license, so it fits research and evaluation use well but requires care before deploying it in a commercial product.

How fast is Aya Vision 32B on Overshoot?

Aya Vision 32B is not currently in Overshoot's live model catalog, so no Overshoot latency figure applies. Models on Overshoot's hosted fast path typically answer in about 200ms; the catalog changes over time, and GET /v1beta/models lists what is currently live.

What is the difference between Aya Vision 32B and Command A Vision?

Aya Vision 32B is a smaller, multilingual-first research model at 32B parameters, while Command A Vision is a larger 112B model built for enterprise document, chart, and table workflows, with a commercial API in addition to open weights.

Related models

• Command A Vision
• Gemma 4 31B
• Qwen3-VL 32B
• ERNIE 4.5 VL