Service Plaza aggregates all services of AI Search Open Platform. You can access Service Plaza to view details of various services without logging on to an Alibaba Cloud account.
Procedure
Go to Service Plaza to view various services provided by AI Search Open Platform.
Service category
Description
Document content parsing
General document parsing service that extracts logical hierarchical structures such as titles and paragraphs from unstructured documents (text, tables, images, etc.) and outputs them in a structured format.
Image content parsing
Image content understanding service: Uses multi-modal LLMs to parse and understand image content and perform text recognition. The parsed text can be used for image retrieval and question-answering scenarios.
Image text recognition service: OCR image text recognition. The recognized text can be used for image retrieval and question-answering scenarios.
Document slice
Provides general text chunking services, supporting the chunking of structured data in HTML, Markdown, and TXT formats based on document paragraphs, text semantics, or specified rules. It also supports extracting code, images, and tables from documents in rich text format.
Text vectorization
OpenSearch text vectorization service-001: Provides multilingual (40+) text embedding service with a maximum input text length of 300 tokens and an output vector dimension of 1536.
OpenSearch universal text vectorization service-002: Provides multilingual (100+) text embedding service with a maximum input text length of 8192 tokens and an output vector dimension of 1024.
OpenSearch text vectorization service-Chinese-001: Provides Chinese text embedding service with a maximum input text length of 1024 tokens and an output vector dimension of 768.
OpenSearch text vectorization service-English-001: Provides English text embedding service with a maximum input text length of 512 tokens and an output vector dimension of 768.
GTE text embedding-multilingual-base: Multilingual (70+) text embedding service with a maximum input text length of 8192 and an output vector dimension of 768.
Qwen3 text embedding-0.6B: Qwen3 series multilingual (100+) text embedding service with a maximum input length of 32k and customizable output vector dimensions from 32 to 1024, with 0.6B parameters.
Multimodal vector
M2-Encoder-multimodal vector model: Chinese-English bilingual multimodal service trained on 6 billion image-text pairs (3 billion Chinese data and 3 billion English data) based on BM-6B. This model supports cross-modal retrieval between text and images (including text searching for images and images searching for text), along with image classification tasks.
M2-Encoder-Large-multimodal vector model: Chinese-English bilingual multimodal service with a larger parameter count of 1B (1 billion parameters) compared to the M2-Encoder model, providing stronger expression capabilities and performance in multimodal task processing.
Text sparse vectorization
Transform text data into sparse vector representations. Sparse vectors require less storage space and are commonly used to represent keywords and term frequency information. Use them with dense vectors for hybrid retrieval to improve retrieval effectiveness.
OpenSearch text sparse vectorization service: Provides multilingual (100+) text vectorization service with a maximum input text length of 8192 tokens.
Query analysis
Based on LLM and NLP capabilities, it can perform intent recognition, similar question expansion, and NL2SQL processing on user queries, effectively improving retrieval and question-answering effectiveness in RAG scenarios.
General query analysis service that uses LLMs to understand user input queries and expand similar questions.
Sorting service
BGE rearrangement model: Provides document scoring service based on the BGE model. It can rank documents according to the relevance between the query and document content, from high to low scores, and output the corresponding scoring results. It supports both Chinese and English languages with a maximum input token length of 512 (Query + Doc length).
OpenSearch text reranking: Trained with multi-industry datasets, it provides high-quality reranking services that can rank documents based on semantic relevance to the query from high to low. It supports both Chinese and English languages with a maximum input token length of 512 (Query + doc length).
Qwen3-Sorting-0.6B: Qwen3 series document reranking service supporting 100+ languages, with a maximum input token length of 32k (Query + doc length) and 0.6B parameters.
Speech recognition
Speech recognition service 001: Provides speech-to-text capabilities that can quickly convert speech content in videos or audio into structured text. This service supports multiple languages.
Video snapshot
Video snapshot service 001: Provides video content extraction capabilities that can capture keyframes from videos. Combined with multimodal embedding services or image parsing capabilities, it enables cross-modal retrieval.
Large model
Qwen3-235B-A22B: The new generation of Qwen LLMs. Based on extensive training, Qwen3 has made breakthroughs in reasoning, instruction following, Agent capabilities, and multilingual support. It supports more than 100 languages and dialects, with powerful multilingual understanding, reasoning, and generation capabilities.
OpenSearch-Qwen-Turbo: Uses qwen-turbo as the base model, with supervised model fine-tuning, enhanced retrieval augmentation, and reduced harmfulness.
Qwen-Turbo: The fastest and most cost-effective model in the Qwen series, suitable for simple tasks. For more information, see Qwen-Turbo.
Qwen-Plus: Balanced in capabilities, with reasoning effectiveness, cost, and speed between Qwen-Max and Qwen-Turbo, suitable for moderately complex tasks. For more information, see Qwen-Plus.
Qwen-Max: The best-performing model in the Qwen series, suitable for complex, multi-step tasks. For more information, see Qwen-Max.
QwQ deep thinking model: A QwQ reasoning model trained based on the Qwen2.5-32B model, significantly improving the model's reasoning ability through reinforcement learning.
DeepSeek-R1: A large language model focused on complex reasoning tasks, with outstanding performance in complex instruction understanding and result accuracy.
DeepSeek-V3: An MoE model with excellent performance in long text, code, mathematics, encyclopedia, and Chinese language capabilities.
DeepSeek-R1-distill-qwen-7b: A model fine-tuned on Qwen-7B using training samples generated by DeepSeek-R1 based on knowledge distillation technology.
DeepSeek-R1-distill-qwen-14b: A model fine-tuned on Qwen-14B using training samples generated by DeepSeek-R1 based on knowledge distillation technology.
Internet search
During the search process, when the private knowledge base cannot provide the corresponding answers, you can expand to search on the Internet to obtain more information, supplement the private knowledge base, and combine with large language models to provide richer answers.
Click View Details in the lower-left corner of the service card to view the Service Name, Service ID, Service Introduction, and Charging method.
Click Experience in the lower-right corner of the service card to go to the Experience Center, where you can test the service and preview its effects. For more information, see Experience center.