Choose alternative products and models for Quanmiao offerings and plan migration for search, video understanding, writing, news aggregation, PPT generation, and intelligent review.
1. Quanmiao API (miaosou_postpaid) — Search / RAG / multimodal APIs
Recommended alternative: Model Studio knowledge bases + Qwen models
| Component | Alternative product / model | Description |
|---|---|---|
| Document parsing | Alibaba Cloud Document Mind (DocMind) | Parses PDFs, images, and tables into Markdown or structured data and outputs semantic chunks. |
| Text embeddings | text-embedding-v4 | Uses 512 dimensions by default and supports 8K tokens. Suitable for general-purpose RAG. |
| Image-text / multimodal embeddings | qwen3-vl-embedding / tongyi-embedding-vision-plus | Supports combined image-text retrieval and cross-modal search. |
| Reranking | qwen3-rerank / gte-rerank-v2 | Reranks the Top-N results from vector retrieval to improve accuracy. |
| Generation models | qwen3.7-plus / qwen3.6-plus | Generates final answers. Supports a 1M context window, function calling, and JSON output. |
| Managed knowledge bases | Model Studio knowledge bases (RAG Knowledge Base) |
Other options
You can also choose other RAG or intelligent search products or open-source solutions based on your business needs. Within the same ecosystem, consider Alibaba Cloud OpenSearch Retrieval Engine Edition or managed Alibaba Cloud Elasticsearch clusters, combined with Qwen models to build your own retrieval pipeline.
2. Video understanding (video_postpaid) — Video analysis and understanding
Recommended alternative: Model Studio multimodal / omni-modal models
| Scenario | Recommended model | Capabilities |
|---|---|---|
| General video understanding | qwen3.7-plus | Supports text, image, and video input. Each video can be up to 2 hours / 2GB, with a 1M context window. |
| Cost-sensitive scenarios | qwen3.6-flash | Also supports videos of up to 2 hours / 2GB, with slightly lower performance at a lower cost. |
| Simultaneous video and audio understanding | qwen3.5-omni-plus / qwen3.5-omni-flash | Supports combined text, image, audio, and video input, with videos of up to 1 hour. Suitable for understanding speech or sound effects in videos. |
| Long-video retrieval / Q&A | Model Studio knowledge bases (audio and video search) | Performs ASR, frame extraction, and plot analysis on long videos, followed by vector-based retrieval. |
Migration path
- Short-video or single-video analysis: Call the OpenAI-compatible API of qwen3.7-plus or qwen3.6-flash directly with a video URL and a prompt.
- Long-video library retrieval:
- Upload videos to OSS.
- Create an audio and video search knowledge base in Model Studio.
- Configure ASR transcription and an embedding model.
- Use the
RetrieveAPI to retrieve segments, then generate answers with qwen3.7-plus.
- If the original API uses a fixed output format, such as tags, summaries, and timestamps, you can reproduce the output schema in the new API with prompts and JSON Mode.
Considerations
- Qwen-Omni models require
stream=True. - Visual and audio information in videos are billed separately. Evaluate the cost of long videos before making calls.
Other options
You can also choose other video understanding products or open-source solutions based on your business needs. Within the same ecosystem, consider the video analysis capabilities of Alibaba Cloud Visual Intelligence Open Platform.
3. Official document assistant (documenttools_postpaid) — Official document writing tools
Recommended alternative: Qwen models
| Approach | Product / model | Description |
|---|---|---|
| Flexible custom implementation | qwen3.7-plus / qwen3.6-plus + prompt engineering | Call models directly and use system prompts to specify the format, style, and terminology of official documents. |
Other options
You can also choose other official document writing products or open-source solutions based on your business needs. Within the same ecosystem, consider office products such as WPS AI official document writing, DingTalk AI Assistant, and Tongyi productivity tools.
4. News aggregation (newshotlist_postpaid) — Trending news aggregation
Original capabilities: Aggregates trending news, generates trending news broadcast lists, and supports channel-based categorization and summarization.
Recommended alternative: Custom aggregation with large language models
| Approach | Product / capability | Description |
|---|---|---|
| Custom implementation | qwen3.7-plus / qwen3.6-plus + built-in web search tool | If official lightweight applications do not meet your business needs, use model web search capabilities to retrieve trending topics and aggregate summaries. |
Other options
You can also choose other news aggregation products or open-source solutions based on your business needs. Within the same ecosystem, use Alibaba Cloud OpenSearch or managed Elasticsearch clusters to build a pipeline that combines crawling, indexing, and LLM summarization. Third-party data options include news APIs from the Toutiao / Douyin Open Platform, Baidu trending search APIs, Weibo trending search APIs, and Tencent News APIs. Combine them with Qwen or other large language models to generate summaries and broadcast lists. For an open-source solution, use an RSS aggregator such as FreshRSS with a crawling framework such as Scrapy to collect content from target sites, then call local or cloud-based models for content aggregation and rewriting.
5. Miaobi (miaobi)
Original capabilities: General-purpose text generation and writing for marketing, office work, creative writing, and other scenarios.
Recommended alternative: Qwen models + AI Miaobi
| Approach | Product / model | Description |
|---|---|---|
| Direct replacement | Model Studio AI Miaobi (AMB) | Shares the name and positioning of the discontinued Miaobi product. Supports professional news, official document, marketing, and office writing. |
| Model-native approach | qwen3.7-plus / qwen3.7-max | Provides the strongest general-purpose text generation capabilities. Suitable for long-form content, creative tasks, code, and complex reasoning. |
| Cost optimization | qwen3.6-plus / qwen3.6-flash | Balances quality and cost. Suitable for high-concurrency scenarios. |
Other options
You can also choose other text generation products or open-source solutions based on your business needs. Within the same ecosystem, consider office and content creation products such as Tongyi productivity tools, DingTalk AI Assistant, and WPS AI.
6. AIPPT (pptgeneration_mixpaid) — AI-powered PPT generation
Original capabilities: Automatically generates PPT content, layouts, and illustrations from a topic or outline.
Recommended alternative: Alibaba Cloud Vibe PPT solution + Model Studio models
| Component | Alternative product / model | Description |
|---|---|---|
| Text outlines | qwen3.6-plus / qwen3.7-plus | Generates a PPT outline and text for each slide based on a topic. |
| Illustration generation | wan2.7-image-pro (Wan) | Generates images that match the topic based on slide content. |
| Custom rendering | Frontend PPT rendering libraries, such as pptxgenjs / python-pptx | For fully customized layouts, render model output into PPTX files yourself. |
Other options
You can also choose other AI PPT products or open-source solutions based on your business needs. Within the same ecosystem, consider office products such as Tongyi productivity PPT tools, WPS AI PPT, and DingTalk AI Assistant.
7. Intelligent review — Content review and proofreading
Original capabilities: Uses Qwen models to meet content review and proofreading needs, including basic document quality, safety and compliance, political sensitivity, enterprise domain knowledge, and image review. Supports custom dictionaries, rule libraries, and trusted information sources.
Recommended alternative: Model Studio intelligent review API + custom Qwen implementation
| Approach | Product / model | Description |
|---|---|---|
| Custom model-based implementation | qwen3.7-plus / qwen3.6-plus + prompt engineering | For general text correction, grammar checks, political sensitivity, and compliance risks, use system prompts to define review dimensions and output formats. Use JSON Mode to return error locations, original text, suggested changes, and supporting grounds. |
| Safety and compliance | Alibaba Cloud Content Moderation (Green) | For safety and compliance checks involving political content, terrorism, pornography, prohibited content, insults, and privacy, call the service independently or combine it with LLM review. |
| Document / image input | Alibaba Cloud Document Mind (DocMind) + qwen3.5-omni-plus | Parse PDFs or image-based documents into text before reviewing them. For image content review, use a vision model or Content Moderation directly. |