Moondream 3.1 9B A2B
Neural Network
Moondream 3.1 is a compact vision-language model by Moondream for image text recognition, descriptions, and visual Q&A.
Specifications
- Input
- text, images, documents
- Output
- text
- Context size
- 32 000 tokens
- Max answer length
- 4 096 tokens
- Reasoning effort
- low, medium, high
- Capabilities
- reasoning
Prompt cost
(per 1M tokens)
Answer cost
(per 1M tokens)
How it works Moondream 3.1 9B A2B?
Moondream 3.1 is a vision-language model by Moondream and a successor to Moondream 3 Preview. It uses a mixture-of-experts architecture with 9 billion parameters, 2 billion of which are active per request. It accepts an image along with a text prompt and processes its content: answers questions about the image, writes descriptions, and performs OCR. Via BotHub, the model is accessible from Russia without a VPN or foreign card, with payment in rubles via Russian cards, charged only for tokens actually used. You can access it via a unified OpenAI-compatible API. With over 250 models available in one window, you can compare responses on the same image and switch without rewriting your integration. Companies have access to contracts, invoices, electronic document management, and an admin panel with limits. It is useful for parsing screenshots and document photos, extracting text from receipts, signs, and packaging, automatically describing catalogs and media libraries, labeling large image datasets, or adding visual prompts to support systems where users send photos.Frequently asked questions about Moondream 3.1 9B A2B
You can use generated results for commercial purposes. You own all rights to the content you create. The only restriction: make sure your prompt does not include copyrighted third-party material. You are responsible for respecting the rights to any input data.
moondream3.1-9b-a2b is a Moondream vision-language model: it accepts text, images, and documents as input and responds with text. It is suitable for describing images, extracting data from scans and screenshots, answering questions about images, and parsing documents. The context window is 32,000 tokens, enough for detailed instructions alongside an image.
The maximum per response is 4096 tokens, which is roughly several pages of text. This is sufficient for a detailed image description, a table with extracted fields, or a comprehensive document analysis. If you need more volume, break the task into parts; the 32,000-token context will maintain the entire conversation.
Yes, reasoning before answering is confirmed for moondream3.1-9b-a2b. The model can analyze a scene step-by-step: first finding objects, checking details, and then formulating a conclusion. This significantly helps when you need to count items in a photo, read fine print, or compare data from a table.
In our data, function calling and strict JSON mode are not noted—this doesn't mean they don't exist, but we cannot guarantee them. A practical approach: request the response in the desired structure directly in the prompt and validate the result on your end. Through the unified BotHub API, it is easy to switch to a model with confirmed function calling.
Images and documents—yes, this is the main scenario: upload a photo, screenshot, scan, or PDF and ask a question about the content. Audio and video are not noted in our data. If you are working with a video, extract the necessary frames and send them as regular images.
There is no information about the quality of specific languages in our data, so we won't make any promises: test the model with your own texts and images; it will take a couple of minutes. BotHub itself is Russian-speaking—interface, support, payment with Russian cards, and access without VPN or foreign cards.