Command A Vision, Explained Simply

Cohere just released Command A Vision, an AI model designed to look at pictures and read text at the same time. It is mainly built for businesses that need to quickly process things like scanned documents, data charts, or photos of tables, turning visual information into plain text.

What it does well

• Chart and table reading: It is good at looking at a graph or a table in an image and explaining the data inside it.
• OCR (Optical Character Recognition): This means it can read printed or written text inside an image and turn it into normal, editable computer text.
• Scene analysis: It can look at a photo and tell you what objects are in it.
• Multilingual: It can read text inside images in English, Portuguese, Italian, French, German, and Spanish.

What it costs and what it can handle

• Context length (memory): It has a 128K token limit. A "token" is a piece of a word, so this means it can remember about 100,000 words of conversation and image data at once.
• Maximum output: It can write up to 8K tokens (roughly 6,000 words) in a single reply.
• Image limits: You can send it up to 20 images in a single message, as long as the total size is under 20MB.

Worth knowing

• It is an "enterprise-focused" tool, meaning it is designed for business and developer use through an API (a way for apps to talk to each other), rather than being a chatbot for casual fun.
• The announcement does not mention a price, so you would need to check Cohere's website to see what they charge to use it.
• The makers note that there are various limitations and best practices for using images, meaning it might not always read messy or unclear images perfectly.