Moondream 3 Preview and 3.1 Now on OpenRouter with Object Detection and Pointing

Key Info

Moondream 3 Preview and Moondream 3.1 from Moondream AI are now available on OpenRouter, served by Prime Intellect. These small vision-language models support object detection, pointing, and captioning directly in chat.

Highlights

• 9B total parameters with 2B active MoE architecture; image + text input with 32K context and optional reasoning
• Chat commands: /detect {"object":"coffee cup"}, /point {"object":"croissant"}, and /caption {"length":"normal"}
• Pricing: $0.30 per million input tokens and $2.50 per million output tokens
• Attach one image and start your message with a command to run detection or pointing