All Products
Search

This Product

    OpenSearch:Experience center

    Document Center

    OpenSearch:Experience center

    Last Updated:Oct 10, 2025

    The Experience Center lets you visually test services such as document parsing, image content extraction, and document segmentation. This helps you quickly assess whether the services meet your business needs.

    Features

    The Experience Center provides the following services:

    Service Category

    Service Description

    Document content parsing

    A general document parsing service that extracts logical structures, such as titles and paragraphs, from unstructured documents containing text, tables, and images. It then outputs the data in a structured format.

    Image content parsing

    Image Content Recognition Service: Uses a multimodal large language model to parse, understand, and recognize text in images. The parsed text can be used for image retrieval and question and answering (Q&A).

    Image Text Recognition Service: An Optical Character Recognition (OCR) service that recognizes text in images. The recognized text can be used for image retrieval and Q&A.

    Document segmentation

    A general text segmentation service that splits structured data in HTML, Markdown, and TXT formats based on document paragraphs, text semantics, or specified rules. This service also supports extracting code, images, and tables from documents as rich text.

    Text embedding

    • OpenSearch Text Embedding Service-001: A multilingual (40+) text embedding service. It supports a maximum input text length of 300 tokens and outputs 1536-dimensional vectors.

    • OpenSearch General Text Embedding Service-002: A multilingual (100+) text embedding service. It supports a maximum input text length of 8192 tokens and outputs 1024-dimensional vectors.

    • OpenSearch Text Embedding Service-Chinese-001: A Chinese text embedding service. It supports a maximum input text length of 1024 tokens and outputs 768-dimensional vectors.

    • OpenSearch Text Embedding Service-English-001: An English text embedding service. It supports a maximum input text length of 512 tokens and outputs 768-dimensional vectors.

    • GTE Text Vector-Multilingual-Base: A multilingual (70+) text embedding service. It supports a maximum input text length of 8192 tokens and outputs 768-dimensional vectors.

    • Qwen3 Text Vector-0.6B: A Qwen3 series multilingual (100+) text embedding service. It supports a maximum input length of 32,000 tokens, outputs 1024-dimensional vectors, and has 0.6 billion parameters.

    Multimodal embedding

    • M2-Encoder-Multimodal Vector Model: A Chinese-English bilingual multimodal service. It is trained on the BM-6B model using a dataset of 6 billion image-text pairs (3 billion Chinese and 3 billion English). This model supports cross-modal retrieval of images and text (including text-to-image and image-to-text search) and image classification tasks.

    • M2-Encoder-Large-Multimodal Vector Model: A Chinese-English bilingual multimodal service. Compared to the M2-Encoder model, it has a larger parameter size of 1 billion (1B). This provides stronger expression and performance in multimodal tasks.

    Text sparse vectorization

    A service that converts text data into sparse vectors. Sparse vectors use less storage space and are often used to represent keywords and term frequencies. They can be combined with dense vectors for hybrid retrieval to improve search results.

    OpenSearch Text Sparse Vectorization Service: A multilingual (100+) text vectorization service. It supports a maximum input text length of 8192 tokens.

    Vector dimension reduction

    embedding-dim-reduction: A vector model fine-tuning service. You can customize and train models for tasks such as vector dimension reduction. This helps reduce high-dimensional vectors to lower dimensions to improve cost-effectiveness without a significant loss in retrieval performance.

    Query analysis

    A query content analysis service that uses large language models (LLMs) and Natural Language Processing (NLP) to analyze user input. It performs intent recognition, alternate query expansion, and natural language to SQL (NL2SQL) processing. This improves retrieval and Q&A performance in retrieval-augmented generation (RAG) scenarios.

    A general query analysis service that uses a large language model for intent recognition and alternate query expansion on user input queries.

    Sorting service

    • BGE rearrange model: A document scoring service based on the BGE model. It sorts documents from high to low based on the relevance between the query and document content, and outputs the corresponding scores. It supports Chinese and English, with a maximum input length of 512 tokens (Query + Doc length).

    • OpenSearch self-developed rearrange model: Trained on datasets from multiple industries, this model provides a high-quality rearranging service. It sorts documents from high to low based on the semantic relevance between the query and the document. It supports Chinese and English, with a maximum input length of 512 tokens (Query + doc length).

    • Qwen3 sorting service-0.6B: A Qwen3 series document rearranging service. It supports over 100 languages, with a maximum input length of 32,000 tokens (Query + doc length), and has 0.6 billion parameters.

    Speech recognition

    Speech Recognition Service 001: Provides speech-to-text capabilities to quickly convert speech from video or audio into structured text. This service supports multiple languages.

    Video snapshot

    Video Snapshot Service 001: Provides video content extraction capabilities that can capture keyframe images from a video. You can combine it with multimodal embedding or image parsing services to enable cross-modal retrieval.

    Large language model

    • Qwen3-235B-A22B: A next-generation Qwen series large language model. With extensive training, Qwen3 has made breakthroughs in reasoning, instruction following, agent capabilities, and multilingual support. It supports over 100 languages and dialects and has powerful multilingual understanding, reasoning, and generation capabilities.

    • OpenSearch-Qwen-Turbo: Based on the Qwen-Turbo large language model, this model is fine-tuned with supervision to enhance retrieval and reduce harmful content.

    • Qwen-Turbo: The fastest and most cost-effective model in the Qwen series. It is suitable for simple jobs. For more information, see the Model list and prices.

    • Qwen-Plus: A balanced model in terms of capability, with inference performance, cost, and speed between those of Qwen-Max and Qwen-Turbo. It is suitable for moderately complex tasks. For more information, see the Model list and prices.

    • Qwen-Max: The best-performing model in the Qwen series. It is suitable for complex, multi-step tasks. For more information, see the Model list and prices.

    • QwQ deep thinking model: A QwQ reasoning model trained on the Qwen2.5-32B model. It uses reinforcement learning to significantly improve its reasoning capabilities.

    • DeepSeek-R1: A large language model that specializes in complex reasoning tasks. It performs well in complex instruction understanding and result accuracy.

    • DeepSeek-V3: A Mixture of Experts (MoE) model that excels in long text, code, math, encyclopedic knowledge, and Chinese language capabilities.

    • DeepSeek-R1-distill-qwen-7b: A model created by fine-tuning Qwen-7B with training samples generated by DeepSeek-R1 using knowledge distillation.

    • DeepSeek-R1-distill-qwen-14b: A model created by fine-tuning Qwen-14B with training samples generated by DeepSeek-R1 using knowledge distillation.

    Web search

    During a search, if the private knowledge base cannot provide an answer, you can expand the search to the web. This gathers more information from the Internet to supplement the private knowledge base and provide richer answers with the help of a large language model.

    Try out the features

    Document parsing

    1. Log on to the AI Search Open Platform console.

    2. In the navigation pane on the left, select Experience Center.

    3. Set Service Category to Document Content Parsing (document-analyze), and select a specific Experience Service.

    4. You can use the sample data provided by the system or upload your own data using Manage Data. Supported file formats include Txt, Pdf, Html, Doc, Docx, Ppt, and Pptx. The file size cannot exceed 20 MB.

      • Upload local file: Uploaded files are automatically purged after seven days. The platform does not store your data long-term.

      • Provide a file URL and the corresponding file type: You can upload multiple URLs. Enter each URL on a new line.

        Note

        Selecting the wrong data format will cause document parsing to fail. Make sure to choose the correct file type for your data.

        image

        Important

        Ensure that you use the web link import feature in compliance with applicable laws and regulations. You must also adhere to the management standards of the target platform and protect the legal rights of rights holders. You are solely responsible for this. As a tool provider, the AI Search Open Platform assumes no liability for your parsing or downloading actions.

    5. If you use your own data, select the pre-uploaded file or URL from the drop-down list.

      image

    6. Click Get Result. The system calls the service to parse the document.

      • Result: Displays the parsing progress and results.

      • Result Source Code: View the response code. Click Copy Code or Download File to download the code locally.

      • Sample Code: View and download sample code for calling the document content parsing service.

        image

    Document segmentation

    1. Log on to the AI Search Open Platform console.

    2. In the navigation pane on the left, select Experience Center.

    3. Set Service Category to Document Segmentation (document-split), and select a specific Experience Service.

    4. You can use the sample data provided by the system or select My Data to enter your own data. Select the correct data format: Txt, Html, or MarkDown.

      Note

      Selecting the wrong data format will cause document parsing to fail. Make sure to choose the correct format for the uploaded data.

    5. Set the maximum segment length. The default value is 300, and the maximum value is 1024. The unit is tokens.

    6. Click Get Result. The system calls the service to segment the document.

      • Result: Displays the segmentation progress and results.

      • Result Source Code: View the response code. Click Copy Code or Download File to download the code locally.

      • Sample Code: View and download sample code for calling the document segmentation service.

    Text or sparse embedding

    1. Log on to the AI Search Open Platform console.

    2. In the navigation pane on the left, select Experience Center.

    3. Set Service Category to Text Embedding (text-embedding), and select a specific Experience Service.

    4. The content types for embedding are document and query.

    5. You can use groups or directly enter JSON text.

      image

    6. Click Get Result to obtain the text embedding result.

      • Result: Displays the embedding result.

        image

      • Result Source Code: View the response code. Click Copy Code or Download File to download the code locally.

      • Sample Code: View and download sample code for calling the text embedding service.

    Multimodal embedding

    1. Log on to the AI Search Open Platform console.

    2. In the navigation pane on the left, select Experience Center.

    3. Set Service Category to Multimodal Embedding (multi-modal-embedding), select a specific Experience Service, and enter text or an image.

      image

      Note

      When you upload a local image for embedding, the image is automatically purged after seven days. The platform does not store your data long-term.

    4. Click Get Result to obtain the multimodal embedding result.

      1. Result: Displays the embedding result.

        image

      2. Result Source Code: View the response code. Click Copy Code or Download File to download the code locally.

      3. Sample Code: View and download sample code for calling the multimodal embedding service.

    Sorting service

    1. Log on to the AI Search Open Platform console.

    2. In the navigation pane on the left, select Experience Center.

    3. Set Service Category to Sorting Service (ranker), and select a specific Experience Service.

    4. You can use the sample data provided by the system or enter your own data.

    5. Enter text in the query field.

      image

    6. Click Get Result. The system calls the sorting service to sort the documents based on the relevance between the query and document content and then outputs the scoring results.

      • Result: Displays the sorting and scoring results.

        image

      • Result Source Code: View the response code. Click Copy Code or Download File to download the code locally.

      • Sample Code: View and download sample code for calling the sorting service.

    Video snapshot

    1. Log on to the AI Search Open Platform console.

    2. In the navigation pane on the left, select Experience Center.

    3. Set Service Category to Video Snapshot (video-snapshot).

    4. You can use the sample data provided by the system or upload your own video data.

    5. Click Get Result. The system calls the video snapshot service to capture keyframe images from the target video.

    Speech recognition

    1. Log on to the AI Search Open Platform console.

    2. In the navigation pane on the left, select Experience Center.

    3. Set Service Category to Speech Recognition (audio-asr).

    4. You can use the sample data provided by the system or upload your own audio data.

    5. Click Get Result. The system calls the speech recognition service to convert the speech content in the target data into structured text.

    Large language model (LLM) service

    1. Log on to the AI Search Open Platform console.

    2. In the navigation pane on the left, select Experience Center.

    3. Set Service Category to Large Model (text-generation), and select a specific Experience Service. You can click image to enable the web search service. The system determines whether to perform a web search based on the user's query.

    4. Enter a query and submit it. The large language model understands the input query and provides a response.

      Important

      All generated content is produced by an artificial intelligence model. We cannot guarantee the accuracy or completeness of the generated content. The content does not represent our attitudes or views.

      The LLM response page displays the input and output token counts for the current Q&A round. You can also delete the current conversation and copy the full text.

    Image content parsing

    1. Log on to the AI Search Open Platform console.

    2. In the navigation pane on the left, select Experience Center.

    3. Set Service Category to Image Content Parsing (image-analyze). In the Experience Service section, select Image Content Recognition or Image Text Recognition.

    4. You can use the sample images provided by the system or upload your own images.

      image

    5. Click Get Result. The system calls the image content parsing service to understand and output the image content or to recognize and output key information from the image.

      • Result: Displays the recognition result.

        image

      • Result Source Code: View the response code. Click Copy Code or Download File to download the code locally.

      • Sample Code: View and download sample code for calling the image content parsing service.

    Query analysis

    1. Log on to the AI Search Open Platform console.

    2. In the navigation pane on the left, select Experience Center.

    3. Set Service Category to Query Analysis (query-analyze).

    4. You can directly enter a Query for query intent recognition. You can also construct a multi-turn conversation in the History area and enter a Query. The model performs query intent recognition based on both the multi-turn conversation and the query.

      Enable the NL2SQL service and select a created service configuration. You can input natural language, and the NL2SQL service converts the natural language query into an SQL statement.

    5. Click Get Result to view the model's performance.

      • Result: Displays the recognition result.

        nl2SQL.png

      • Result Source Code: View the response code. Click Copy Code or Download File to download the code locally.

      • Sample Code: View and download sample code for calling the query analysis service.

    Vector fine-tuning

    1. Log on to the AI Search Open Platform console.

    2. In the navigation pane on the left, select Experience Center.

    3. Set Service Category to Vector Dimension Reduction (embedding-dim-reduction).

    4. Select the model name, which is the model you fine-tuned on your business data. Fill in the Output Vector Dimension. The output vector dimension must be less than or equal to the dimension of the vector field selected during model training. Then, enter the original vector.

    5. Click Get Result to view the model's performance.

      For more information about how to train a dimension reduction model, see Service customization.

    Web search

    You can use web search in the following two ways:

    • Call the web search service directly.

    • Enable web search when using an LLM model.

    1. Log on to the AI Search Open Platform console.

    2. Select the destination region and switch to the AI Search Open Platform.

    3. In the navigation pane on the left, select Experience Center.

    4. Set Service Category to Web Search Service (web-search).

    5. Enter a query, such as "what to do in Hangzhou", and view the results.