The AI Search Open Platform is compatible with the OpenAI API. You can call its services using the OpenAI SDK.
Two categories of services are available: Text embedding and Content generation.
Text embedding
Use these models to convert text into vector representations for semantic search and similarity tasks.
| Model | ID | Languages supported | Max input (tokens) | Output dimensions |
|---|---|---|---|---|
| OpenSearch Text Embedding Service-001 | ops-text-embedding-001 | 40+ | 300 | 1536 |
| OpenSearch General Text Embedding Service-002 (recommended) | ops-text-embedding-002 | 100+ | 8192 | 1024 |
| OpenSearch Text Embedding Service-Chinese-001 | ops-text-embedding-zh-001 | Chinese only | 1024 | 768 |
| OpenSearch Text Embedding Service-English-001 | ops-text-embedding-en-001 | English only | 512 | 768 |
Not sure which to pick? Start with OpenSearch General Text Embedding Service-002 (ops-text-embedding-002). It supports the widest range of languages and accepts the longest input, making it the best general-purpose choice. Use the language-specific models when you need to optimize for a single language.
For API usage details, see Text Embedding.
Content generation
Use these models for retrieval-augmented generation (RAG) and other language tasks.
OpenSearch-tuned model
| Model | ID | Description |
|---|---|---|
| OpenSearch-Qwen-Turbo | ops-qwen-turbo | Built on qwen-turbo with supervised fine-tuning and retrieval-augmented generation. Optimized for safety and relevance. |
Qwen series
| Model | ID | Best for |
|---|---|---|
| Qwen-Turbo | qwen-turbo | Simple tasks. Fastest and most cost-effective in the Qwen series. |
| Qwen-Plus | qwen-plus | Medium-complexity tasks. Balances inference quality, cost, and speed between Qwen-Max and Qwen-Turbo. |
| Qwen-Max | qwen-max | Complex, multi-step tasks. Highest-performing model in the Qwen series. |
For full model specifications, see the Model List for Qwen-Turbo, Qwen-Plus, and Qwen-Max.
DeepSeek series
| Model | ID | Description |
|---|---|---|
| DeepSeek-R1 | deepseek-r1 | Optimized for complex reasoning tasks. Excels at understanding complex instructions and delivering accurate results. |
| DeepSeek-V3 | deepseek-v3 | Mixture of Experts (MoE) architecture. Strong in long-context understanding, coding, math, encyclopedic knowledge, and Chinese language tasks. |
| DeepSeek-R1-distill-qwen-7b | deepseek-r1-distill-qwen-7b | Qwen-7B model fine-tuned on DeepSeek-R1 training samples via knowledge distillation. |
| DeepSeek-R1-distill-qwen-14b | deepseek-r1-distill-qwen-14b | Qwen-14B model fine-tuned on DeepSeek-R1 training samples via knowledge distillation. |
For API usage details, see Content Generation.