What is moonshot-v1-vision-preview
moonshot-v1-vision-preview is a multimodal image understanding model launched by Moonshot AI. The model improves the multimodal capabilities of the moonshot-v1 model series and has powerful image recognition capabilities, able to accurately distinguish complex details, such as similar images of blueberry muffins and Chihuahuas. In text recognition, the model performs excellently and can accurately recognize messy handwritten content, such as receipts and express delivery forms. moonshot-v1-vision-preview can analyze data in images, such as subject scores in bar charts, and evaluate charts from an aesthetic perspective. The model is based on API calls and supports features such as multi-turn conversations and streaming output.
Main features of moonshot-v1-vision-preview
- Image recognition: Accurately recognizes complex details and subtle differences in images. Even for objects with high similarity that are difficult for the human eye to distinguish, such as images of blueberry muffins and Chihuahuas, the model can accurately distinguish and recognize them.
- OCR text recognition capability: Performs outstandingly in OCR text recognition and image understanding scenarios, more accurately than ordinary file scanning and OCR recognition software. It can recognize messy handwritten content in documents such as receipts and express delivery forms, and accurately extract text information.
- Image data extraction and analysis: Accurately recognizes data information in images, such as subject names and score values in bar charts, and performs comparative data analysis. It can also recognize aesthetic elements such as image style, format, and color, and evaluate images from an aesthetic perspective.
- API calls: Based on API calls, users can integrate the model into their own applications.
Project address of moonshot-v1-vision-preview
- Project official website：https://platform.moonshot.cn/docs/guide/use-kimi-vision-model
- Product experience：Kimi Intelligent Assistant
Usage instructions for moonshot-v1-vision-preview
- The features supported by the model include:Multi-turn conversations, streaming output, tool calls, JSON Mode, Partial Mode
- The following features are not yet supported or are partially supported
- Web search: Not supported
- Context Caching: Creating a Context Cache with image content is not supported. Using an already successfully created Cache to call the Vision model is supported.
- Images in URL format: Not supported. Currently only image content encoded in base64 is supported.
Model pricing of moonshot-v1-vision-preview
- moonshot-v1-8k-vision-preview：1M tokens ￥12.00
- moonshot-v1-32k-vision-preview：1M tokens ￥24.00
- moonshot-v1-128k-vision-preview：1M tokens ￥60.00
Application scenarios of moonshot-v1-vision-preview
- Content moderation and classification: Automatically recognizes and classifies image content, suitable for product management on e-commerce platforms, animal and plant recognition in scientific research and education, and platform content moderation.
- Document and data processing: Efficiently extracts text information from documents and tables, suitable for document processing of contracts and invoices, and data analysis of grade sheets and financial statements.
- Medical and industrial applications: Assists medical imaging analysis and improves diagnostic accuracy; performs defect detection on industrial products and improves production quality.
- Intelligent interactive services: In the fields of intelligent customer service, educational assistance, and smart homes, provides smarter and more convenient services based on multimodal interaction.
- Aesthetics and design evaluation: Analyzes images from an aesthetic perspective, provides improvement suggestions for advertising and web design, assists artistic creation, and enhances visual effects.