阿里云 DashScope(通义千问)API
阿里云 DashScope(通义千问)API
DashScope API is Alibaba Cloud's LLM platform — Qwen LLM, Qwen-VL multimodal, text embeddings, TTS, and other Alibaba self-developed models.
Qwen series approaches GPT-4 on Chinese benchmarks
English scenarios weaker than GPT-4 / Claude
Enable service at dashscope.aliyun.com for API-KEY. POST /api/v1/services/aigc/text-generation/generation invokes qwen-max.
Uptime · 30-day window
About this API
DashScope (Chinese: Lingji) is the LLM platform launched by Alibaba Cloud in 2023, benchmarking OpenAI Platform. Main offering is Alibaba's self-developed Qwen series: Qwen-Max (flagship vs. GPT-4), Qwen-Plus (mid-tier), Qwen-Turbo (low cost), Qwen-VL (multimodal vision), Qwen-Audio (audio), Qwen-Coder (code), text-embedding-v3 (vectorization). Alibaba also open-sourced Qwen extensively (7B to 72B multiple sizes fully open-sourced), high download volume on HuggingFace; one of the strongest Chinese open-source LLMs. DashScope API design is OpenAI-compatible-ish (function calling, JSON mode, etc.); low switching cost for developers. Default choice for Chinese developers building Chinese LLM apps. OpenAI still better for overseas / English scenarios.
What you can build
- 1Chinese LLM apps (chatbots, content generation)
- 2Multimodal understanding (Qwen-VL visual QA)
- 3Chinese RAG apps (high embedding accuracy)
- 4Outbound products needing mainland China region
Strengths & limitations
Strengths
- Qwen series approaches GPT-4 on Chinese benchmarks
- Mainland China region deployment (compliance-friendly)
- Open-source versions (Qwen 7B/14B/72B) downloadable for local runs
- Cheaper than OpenAI
Limitations
- English scenarios weaker than GPT-4 / Claude
- Ecosystem weaker than OpenAI
- Requires Alibaba Cloud account
Example request
curl https://dashscope.aliyun.com/<endpoint> \
-H "Authorization: Bearer $API_KEY"
# Some providers use X-Api-Key instead — verify in the docs.Getting started
Enable service at dashscope.aliyun.com for API-KEY. POST /api/v1/services/aigc/text-generation/generation invokes qwen-max.
FAQ
Is Qwen stronger than GPT-4 on Chinese benchmarks?+
On some Chinese-specific benchmarks Qwen-Max slightly beats GPT-4; overall on par. Qwen is more sensitive to Chinese cultural context.
Can the open-source version run locally?+
Yes. Download Qwen2.5 series from HuggingFace; ollama / vLLM both support. Watch license terms for commercial use.
Technical details
- Auth type
- api_key
- Pricing
- paid
- Rate limit
- 默认 RPM/TPM 按模型不同;申请提高
- Protocols
- REST
- SDKs
- python, javascript, java