All Products
Search
Document Center

OpenSearch:Release notes

Last Updated:Apr 01, 2026

This page tracks feature releases and model updates for AI Search Open Platform.

2025

June

TypeFeatureDescriptionRelease dateReference
NewSpeech recognitionCall the speech recognition API to convert speech from videos or audio files into structured text. Supports meeting transcription, video content retrieval, and online customer service.2025-06-26Speech recognition
NewVideo snapshotCall the video snapshot API to extract keyframes from videos. Combine with optical character recognition (OCR), image parsing, or multimodal embedding for deep video content analysis.2025-06-26Video snapshot
Updateops-qwen3-embedding-0.6b added to text embeddingNew multilingual text embedding model from the Qwen3 series. Key specs: 100+ languages supported, 32k max input length, 32–1024 customizable output dimensions, 0.6B parameters.2025-06-26Text embedding
Updateops-qwen3-reranker-0.6b added to re-rank serviceNew re-ranking model from the Qwen3 series. Key specs: 100+ languages supported, 32k max input token length (query + document combined), 0.6B parameters.2025-06-26Re-rank service
Updateops-gme-qwen2-vl-2b-instruct added to multimodal embedding serviceNew multimodal embedding model trained on the Qwen2-VL multimodal large language model (MLLM). Handles text-only, image-only, and mixed text-image inputs.2025-06-26Multimodal embedding
NewMultimodal embeddingMultimodal embedding service trained on the Qwen2-VL MLLM. Supports single-modal and multimodal combination inputs, including text, images, and mixed data.2025-06-04Multimodal embedding

April

TypeFeatureDescriptionRelease dateReference
UpdateQwen3-235B-A22B added to LLM serviceNew large language model (LLM) from the Qwen3 series. Key specs: 100+ languages and dialects supported, with improvements in reasoning, instruction following, and agent capability.2025-04-29Content generation

March

TypeFeatureDescriptionRelease dateReference
NewWeb searchCall the web search API independently or use it within LLM-based conversational search workflows.2025-03-20Internet search
UpdateQwQ deep thinking modelReasoning model trained on Qwen2.5-32B with reinforcement learning. Performance on math and coding benchmarks (AIME 24/25, LiveCodeBench) and general benchmarks (IFEval, LiveBench) reaches the level of the DeepSeek-R1 full version.2025-03-20Content generation
UpdateModel deploymentModelScope is now supported as a model source. Newly supported model types: multimodal embedding models and re-ranking models.2025-03-13Model deployment
NewService developmentAI Search Open Platform is now integrated with Data Science Workshop (DSW) of Platform for AI (PAI). Use Notebook in DSW to develop and run services.2025-03-13Service development

February

TypeFeatureDescriptionRelease dateReference
NewDeepSeek modelsAll DeepSeek models are now available, including DeepSeek R1/V3 and 7B/14B distilled models. Call them using the OpenAI-compatible API.2025-02-14Model list

January

TypeFeatureDescriptionRelease dateReference
NewModel deploymentDeploy models independently in AI Search Open Platform for inference services with higher concurrency and lower latency.2025-01-07Model deployment

2024

October

TypeFeatureDescriptionRelease dateReference
NewVector dimensionality reduction serviceTrain custom vector dimensionality reduction models using your own vector data.2024-10-23Vector dimensionality reduction service

September

TypeFeatureDescriptionRelease dateReference
NewRe-ranking modelSelf-developed re-ranking model trained on multi-industry datasets. Supports multiple languages and outperforms open-source bge-rerank series models across 8 retrieval tasks on the Chinese MTEB benchmark.2024-09-12Re-rank service
NewNL2SQL added to the query analysis serviceNatural language to SQL (NL2SQL) capability added to the query analysis service. Configure and call the NL2SQL service to convert natural language queries into SQL statements.2024-09-12Configure the NL2SQL feature and Query analysis
NewMultimodal data processing scenarioNew scenario added to the scenario center for text and image parsing and embedding.2024-09-12Parse and embedding multimodal data
NewLlamaIndex code architecture added to the RAG scenarioLlamaIndex code architecture added to the scenario center for building retrieval-augmented generation (RAG)-based conversational search applications.2024-09-12Build a RAG-based conversational search application

August

TypeFeatureDescriptionRelease dateReference
OptimizedOCR optimization in image parsingOCR performance in image parsing improved by approximately 40%.2024-08Image content extraction

July

TypeFeatureDescriptionRelease dateReference
NewEvaluation managementEvaluate the effectiveness of your RAG pipeline. Upload an evaluation dataset and get LLM-based evaluation results.2024-07-08Manage evaluation tasks
NewData upload in the experience centerUpload your own data to the experience center to test services with real content.2024-07-08Experience center
NewOpenAI SDK compatibilityCall AI Search Open Platform services using the OpenAI SDK. The platform is compatible with OpenAI interfaces.2024-07-08Supported service list
NewLangChain code architecture added to the RAG scenarioLangChain code architecture added to the scenario center for building RAG-based conversational search applications.2024-07-08Build a RAG-based conversational search application
NewQuery analysis serviceAnalyze queries using LLMs and NLP to understand user intent, generate similar questions, and convert natural language to SQL. Improves conversational search performance in RAG scenarios.2024-07-08Query analysis
NewImage parsing serviceParse image content using multimodal LLMs, including text extracted from images, for use in image retrieval and conversational search.2024-07-08Image content extraction
NewQuery analysis service experienceQuery analysis service added to the experience center. Test intent detection and similar query generation based on historical messages.2024-07-30Experience center
NewImage parsing in RAG scenariosImage parsing in documents is now supported in RAG scenarios. Combined with the query analysis service, run conversational search on rewritten queries.2024-07-30Build a RAG-based conversational search application
NewVPC access regionsVPC access to services now supported in the China (Shanghai), China (Hangzhou), China (Shenzhen), China (Beijing), China (Qingdao), and China (Zhangjiakou) regions.2024-07-30Query service endpoint

June

TypeFeatureDescriptionRelease dateReference
NewRAG-based conversational search applicationEnd-to-end RAG solution for knowledge base conversational search, covering data preprocessing, data retrieval, and output generation.2024-06-18Build a RAG-based conversational search application

May

TypeFeatureDescriptionRelease dateReference
NewLLMLLM service added, including Qwen models and fine-tuned RAG-specific LLMs.2024-05-21Service overview
NewRe-ranking serviceRe-ranking service added for general document scoring. Ranks documents by query-document relevance in descending order and returns scores.2024-05-21Re-rank service
NewSparse text embedding serviceSparse text embedding service added. Converts text into sparse vectors.2024-05-21Sparse text embedding
NewText embedding serviceText embedding service added. Converts text into dense vectors.2024-05-21Text embedding
NewDocument chunking serviceDocument chunking service added. Split HTML, Markdown, and TXT documents by paragraph structure, semantic meaning, or custom rules. Extract code, images, and tables from rich-text content.2024-05-21Document chunking
NewDocument parsing serviceDocument parsing service added. Extract logical structure and content from unstructured documents, including titles, paragraphs, text, tables, and images, in structured output format.2024-05-21Document content parsing
NewService experienceService experience added to the console. Try text parsing, chunking, embedding, re-ranking, LLM, and other services directly in the browser.2024-05-21Experience center