moondream3-preview

by moondream

Compact visual reasoning model for fast image QA and scene captions

276Kdl/month

Compact VLMparams

HuggingFace Use in Pipeline

Identifiers

Model ID

moondream/moondream3-preview

Feature URI

mixpeek://image_extractor@v1/moondream3_preview_v1

Overview

Moondream3 Preview is a compact image-text model from Moondream focused on visual question answering, captioning, and deployable visual reasoning. It continues the Moondream line's emphasis on small-model ergonomics while keeping enough visual reasoning quality for production perception pipelines.

On Mixpeek, Moondream3 is a useful second-stage model after cheap embedding retrieval. Use it to caption candidate images, answer bounded visual questions, or extract concise observations that an agent can cite.

Architecture

Image-text-to-text model exposed through Hugging Face Transformers custom code. It supports caption generation, visual question answering, and streaming output for interactive applications.

Mixpeek SDK Integration

import { Mixpeek } from "mixpeek";

const mx = new Mixpeek({ apiKey: "API_KEY" });

await mx.collections.ingest({
  collection_id: "image-library",
  source: { url: "s3://assets/images/" },
  feature_extractors: [{
    feature: "scene_caption",
    model: "moondream/moondream3-preview",
    params: { caption_length: "short" }
  }]
});