Document Center

All Products

Document Center

OpenSearch:Release notes

Last Updated:Aug 22, 2026

This topic describes product features and documentation updates for the AI Search Open Platform.

2026

April

Category	Feature Name	Description	Release date	References
New	Agentic Memory service	Adds the Agentic Memory service to provide AI agents with structured long-term memory management. It supports create, read, update, delete, and intelligent retrieval of memories.	2026-04	Agentic Memory service
New	Agentic Memory tutorial	Adds a best practice guide on using the Agentic Memory API to enhance OpenClaw with persistent long-term memory.	2026-04	Enhance OpenClaw long-term memory using the Agentic Memory API
New	Video segmentation	Adds video segmentation service that performs semantic understanding and scene analysis to automatically split videos into segments. It outputs time ranges, keyframe screenshots, and captions for each segment. This is useful for video indexing, content understanding, and multimodal RAG scenarios.	2026-04	Video segmentation
New	Video summarization	Adds video summarization service that analyzes video content semantically to auto-generate titles, summaries, and tags. It also outputs enhanced captions and metadata for each segment. This is useful for video understanding, auto-annotation, and multimodal RAG scenarios.	2026-04	Video summarization
Update	Billing methods and items	Updates billing documentation to include pricing details for the Agentic Memory service.	2026-04	Billing methods and items

2025

December

Category	Feature Name	Description	Release date	References
New	Subject identification	Detects and identifies one or more primary objects in images. This applies to typical computer vision scenarios such as smart surveillance, autonomous driving, and image retrieval.	2025-12-19	Subject identification
New	Multimodal ranking	Provides image relevance ranking. In RAG and multimodal search scenarios, this service finds highly relevant content and returns results in ranked order, improving retrieval accuracy and large language model (LLM) generation quality.	2025-12-19	Multimodal ranking
Update	Multimodal embedding	Adds: ops-mm-embedding-v1-2b: A multimodal embedding model developed by Alibaba Cloud OpenSearch-AI team, fine-tuned from Qwen2-VL 2B Instruct. It supports text, image, video, and multimodal inputs, encoding them into unified semantic vectors for cross-modal retrieval and understanding tasks. ops-mm-embedding-v1-7b: A multimodal embedding model developed by Alibaba Cloud OpenSearch-AI team, fine-tuned from Qwen2-VL 7B Instruct. It supports text, image, video, and multimodal inputs, encoding them into unified semantic vectors for cross-modal retrieval and understanding tasks.	2025-12-19	Multimodal embedding
Update	Document content parsing	Adds: Document Content Parsing Service 002: Uses PaddleOCR-VL / MinerU2.5-vlm models to parse unstructured documents like PDFs and images. It excels at recognizing complex elements such as tables, formulas, and charts, and offers fast inference speed.	2025-12-19	Document content parsing

June

Category	Feature Name	Description	Release date	References
New	Speech recognition	The AI Search Open Platform supports calling speech recognition via API to convert spoken content in audio or video into structured text. Use cases include meeting transcripts, video search, and online customer service.	2025-06-26	Speech recognition
New	Video snapshot	The AI Search Open Platform supports calling video snapshot via API to extract keyframes. Combine this with OCR, image parsing, or multimodal embedding services for deep video content analysis and structured processing.	2025-06-26	Video snapshot
Update	Add ops-qwen3-embedding-0.6b to text embedding	Qwen3 series multilingual (100+) text embedding service. Supports input length up to 32k tokens. Output vector dimensions are customizable from 32 to 1024. Model size: 0.6B parameters.	2025-06-26	Text embedding
Update	Add ops-qwen3-reranker-0.6b to ranking service	Qwen3 series document reranking service supporting 100+ languages. Maximum input token length (query + document): 32k. Model size: 0.6B parameters.	2025-06-26	Ranking service
Update	Add ops-gme-qwen2-vl-2b-instruct to multimodal embedding service	Multimodal embedding service trained on Qwen2-VL multimodal large language models (MLLMs). Supports single-modality and multimodal inputs, efficiently handling text, images, and combined data types.	2025-06-26	Multimodal embedding
New	Multimodal embedding	Multimodal embedding (multi-modal-embedding) service trained on Qwen2-VL multimodal large language models (MLLMs). Supports single-modality and multimodal inputs, efficiently handling text, images, and combined data types.	2025-06-04	Multimodal embedding

April

Category	Feature Name	Description	Release date	References
Update	Add Qwen3-235B-A22B to large model service	The new Qwen3 series large language model achieves breakthroughs in reasoning, instruction following, agent capabilities, and multilingual support after extensive training. It supports over 100 languages and dialects with strong multilingual understanding, reasoning, and generation abilities.	2025-04-29	Content generation service

March

Category	Feature Name	Description	Release date	References
New	Web search	Releases web search functionality. Call the web search API directly or use it during LLM response generation.	2025-03-20	Web search
Update	Add QwQ deep thinking model to LLMs	QwQ is a reasoning model trained from Qwen2.5-32B. Reinforcement learning significantly boosts its reasoning ability. It matches DeepSeek-R1 full-performance levels on core benchmarks (AIME 24/25, LiveCodeBench) and some general benchmarks (IFEval, LiveBench).	2025-03-20	Content generation service
Update	Model deployment	Adds ModelScope as a model source. Adds multimodal embedding models and ranking models as supported model types.	2025-03-13	Model deployment
New	Service development	The AI Search Open Platform integrates PAI DSW. Developers can now complete service development and execution directly in the platform using Notebook.	2025-03-13	Service development

February

Category	Feature Name	Description	Release date	References
New	DeepSeek models	The AI Search Open Platform supports the full DeepSeek model series (including R1/V3 and 7B/14B distilled versions) with OpenAI API-compatible calling.	2025-02-14	Model list

January

Category	Feature Name	Description	Release date	References
New	Model deployment	On-premises deployment of models on the AI Search Open Platform provides higher concurrency and lower latency inference services.	2025-01-07	Model deployment

2024

October

Category	Feature Name	Description	Release date	References
New	Vector dimensionality reduction service	Supports custom training of vector dimensionality reduction models using user-provided vector data.	2024-10-23	Vector dimensionality reduction service

September

Category	Feature	Description	Release date	References
New	Add rerank model	Adds a proprietary rerank model trained on multi-industry datasets with multilingual support. It outperforms open-source bge-rerank series models on 8 Chinese C-MTEB retrieval tasks.	2024-09-12	Ranking service
New	Add NL2SQL to query analysis service	Adds NL2SQL capability to the query analysis service, supporting NL2SQL service configuration and invocation.	2024-09-12	NL2SQL service configuration Query analysis
New	Add multimodal data processing scenario	Adds a multimodal data processing scenario to the scenario center, supporting text and image parsing and vectorization.	2024-09-12	Multimodal data parsing and vectorization
New	Add Llamaindex code architecture to RAG scenario	Adds Llamaindex code architecture to the RAG-based knowledge base Q&A scenario in the scenario center.	2024-09-12	Build a knowledge base Q&A system using RAG

August

Category	Feature Name	Description	Release date	References
Optimization	OCR optimization for image parsing	Optimizes the image parsing OCR service, improving performance by approximately 40%.	2024-08	Image content extraction

July

Category	Feature Name	Description	Release date	References
New	Evaluation management	Use the evaluation module to assess RAG pipeline performance. After uploading an evaluation dataset, the system generates evaluation results using large models.	2024-07-08	Evaluation task management
New	Upload data in Experience Center	The Experience Center now supports uploading your own data for service testing.	2024-07-08	Service overview and experience
New	OpenAI SDK compatibility	The AI Search Open Platform is compatible with OpenAI APIs. Developers can use the OpenAI SDK to call services provided by the AI Search Open Platform.	2024-07-08	Supported services list
New	Add Langchain code architecture to RAG scenario	Adds Langchain code architecture to the RAG-based knowledge base Q&A scenario in the scenario center.	2024-07-08	Build a knowledge base Q&A system using RAG
New	Query analysis service	Provides query content analysis using large language models and NLP capabilities. It preprocesses user queries, performs intent recognition, and expands queries with alternate questions to improve RAG retrieval and Q&A effectiveness.	2024-07-08	Query analysis
Add	Image parsing service	Adds image content parsing service. Use multimodal large models to understand image content and recognize text, or use OCR for text extraction. The parsed text supports image retrieval and Q&A scenarios.	2024-07-08	Image content extraction
New	Query analysis service experience	Adds query analysis service to the Experience Center. It provides intent detection and alternate queries based on historical messages and current queries.	2024-07-30	Service overview and experience
New	Add image parsing capability to RAG scenario	Adds image parsing capability for documents in RAG scenarios. Also enhances the query analysis service to support follow-up Q&A based on rewritten queries.	2024-07-30	Build a knowledge base Q&A system using RAG
New	Add VPC access regions	Supports VPC access in Shanghai, Hangzhou, Shenzhen, Beijing, Qingdao, and Zhangjiakou regions.	2024-07-30	Get service access endpoints

June

Category	Feature Name	Description	Release date	References
New	Build a knowledge base Q&A system using RAG	For knowledge base Q&A scenarios, the AI Search Open Platform provides a complete RAG development pipeline. The pipeline includes three modules: data preprocessing, retrieval service, and Q&A summarization generation.	2024-06-18	Build a knowledge base Q&A system using RAG

May

Category	Feature Name	Description	Release date	References
New	Large model service	Adds large model service, including Qwen and fine-tuned RAG-specific large models.	2024-05-21	Large model service
New	Ranking service	Adds ranking service that scores documents based on query-document relevance. It sorts documents from highest to lowest score and returns the scores.	2024-05-21	Ranking service
New	Text sparse vector service	Adds text sparse vector service that converts text data into sparse vector representations.	2024-05-21	Text sparse vector
New	Text embedding service	Adds text embedding service that converts text data into dense vector representations.	2024-05-21	Text embedding
New	Document segmentation service	Adds document segmentation service with general text segmentation strategies. It splits structured data in HTML, Markdown, and TXT formats based on document structure, semantics, or custom rules. It also extracts code, images, and tables in rich text format.	2024-05-21	Document segmentation
New	Document parsing service	Adds document parsing service that extracts logical structures (such as titles and sections) and content (such as text, tables, and images) from unstructured documents, outputting them in a structured format.	2024-05-21	Document parsing
New	Service experience	Adds service experience capability in the console, supporting text parsing, segmentation, embedding, ranking, and large model services.	2024-05-21	Service overview and experience