Alibaba Model Studio (DashScope)

Getting credentials

Endpoints

API surfaces

textstreamingembeddingsbatch

Quirks

• China base: https://dashscope.aliyuncs.com/compatible-mode/v1. Soft switches /think and /no_think appended to prompts work on open-source hybrid Qwen3 models and qwen-plus-2025-04-28+ (most recent instruction wins). Open-source hybrid models (qwen3-235b-a22b, qwen3-32b) support streaming only when enable_thinking=true — non-streaming calls error. Thinking-only variants (qwen3-next-80b-a3b-thinking, qwq-plus, kimi-k2-thinking) cannot disable. Regional OpenAI-compatible bases like https://{workspace}.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1 also exist; native DashScope endpoint .../api/v1/services/aigc/text-generation/generation with X-DashScope-SSE: enable. (docs)
• Pay-as-you-go regions: China dashscope.aliyuncs.com, Singapore dashscope-intl, Virginia dashscope-us — each with /compatible-mode/v1 (OpenAI), /apps/anthropic (Anthropic), and native /api/v1. Workspace-dedicated domains https://{WorkspaceId}.{region}.maas.aliyuncs.com are recommended for production; rate-limited trial domains exist (trial.cn-beijing / trial.ap-southeast-1). (docs)
• Temporary API keys: POST /api/v1/tokens (Bearer with permanent key) returns short-lived st- tokens (default 60s, max 1800s) inheriting the parent key's permissions — for browser/mobile use. Alibaba Cloud AccessKey/STS is NOT accepted on inference endpoints. (docs)
• Savings/resource plans are billing constructs on the same keys and endpoints: AI General-purpose Savings Plan (from $150/mo, discounts up to 32%/47%, no discount for DeepSeek/Kimi/GLM/MiniMax) and LLM Savings Plan prepaid credits (no discount). QwenCloud (qwencloud.com) is a separate-account console using these same endpoints and key formats; Bailian is the China brand of Model Studio. (docs)

Models offered