兔子API
provider
Alibaba Qwen
Direct answerFor direct Model Studio access, copy the API host for the target region and workspace and use a key from the same scope. New pay-as-you-go keys often start with sk-ws-, plan keys with sk-sp-, while this gateway uses its own token.
Updated · Reviewed
Qwen versus Model Studio
Qwen is Alibaba's model family; Model Studio hosts Qwen together with third-party models, so not every marketplace entry belongs to Qwen. This gateway page declares only /v1/chat/completions; native DashScope, Anthropic-compatible, and media APIs are not gateway endpoints by implication.
Direct access versus this gateway
Direct Model Studio access uses the API host and key for a specific region and workspace. A Beijing OpenAI-compatible host looks like https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1; new integrations should copy the host generated by the console. Gateway calls use this site's base URL, Bearer token, and an enabled gateway model name. Never send a Model Studio key to the gateway endpoint.
Keys, regions, and model values
New pay-as-you-go keys often start with sk-ws-, older sk- keys can remain valid, and Token Plan or Coding Plan uses a separate sk-sp- key. A key's plaintext is displayed once and its workspace, IP, or model scope may be restricted. Region, key, host, and model availability must match, and requests use the exact model ID from the official model page rather than a console display title.
Protocols, streams, and tools
OpenAI-compatible Chat streams SSE and function tools use JSON Schema. The application executes functions and validates arguments. Protocol support remains model-specific; Qwen-Audio is an explicit native-DashScope-only exception. Through this gateway, fields also depend on the active channel, and provider extensions may not pass through every adapter.
Limits and cost safety
Limits aggregate at the primary account, so RAM users, workspaces, and multiple keys do not automatically receive independent capacity. Burst RPS/TPS can return 429 even below minute totals. Add queues, concurrency caps, exponential backoff, and usage alerts, and match plan keys with plan hosts to avoid unintended pay-as-you-go billing. Use live regional pricing and quota pages.
Reviewed provider catalog
Latest models by capability
These are provider-catalog models, not a promise of availability on this site. Confirm callable IDs, endpoints, and pricing in the model marketplace.
Text, reasoning, and code
-
qwen3.8-maxStableThe flagship Qwen 3.8 model for complex reasoning, coding, and agentic tasks.
-
qwen3.7-plusStableThe balanced general-purpose tier in the Qwen family.
-
qwen3.7-flashStableA lightweight Qwen model for low-latency, high-throughput workloads.
Image generation and editing
-
qwen-image-3.0-proStableQwen's latest professional image generation and editing model.
Video generation and editing
-
wan2.7-t2vStableThe Wan 2.7 text-to-video model with synchronized audio and multi-shot storytelling.
-
wan2.7-i2vStableThe Wan 2.7 image-to-video model supporting first frame, first/last frames, and continuation.
Speech and audio
-
qwen-audio-3.0-tts-plusStableThe high-quality speech synthesis model in the Qwen Audio 3.0 family.
-
qwen3.5-omni-plus-realtimeStableA Qwen Omni model for realtime multimodal and voice interactions.
Embeddings, retrieval, and reranking
-
text-embedding-v4StableThe current Model Studio text embedding model for semantic search and RAG.
-
qwen3-rerankStableThe Qwen 3 reranking model for improving retrieval result order.
Use cases
- General and deep reasoning
- Coding and agents
- Multimodal and media generation
API protocols
/v1/chat/completions
FAQ
What base URL should an OpenAI SDK use for Model Studio?
Copy the API host for the target region and workspace. A Beijing workspace host looks like https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1. Host, key, and model region must match; gateway calls instead use this site's base URL and Bearer token.
What is the difference between sk-ws-, sk-, and sk-sp- keys?
New pay-as-you-go keys often start with sk-ws-, while older sk- keys may remain valid. Token Plan and Coding Plan use separate sk-sp- keys. Plaintext is shown only once, and plan keys must not be mixed with pay-as-you-go hosts.
Why do I get 429 while my per-minute totals look safe?
Model Studio can enforce burst RPS/TPS as well as RPM/TPM, and limits aggregate across RAM users, workspaces, and keys under the primary account. Smooth traffic with a queue, cap concurrency, back off exponentially, and use a verified fallback model.
Do all Qwen models support the OpenAI-compatible API?
No. Many text models support OpenAI Chat and function tools, but the official guide identifies Qwen-Audio as DashScope-native only. Check the exact model page for vision, audio, tools, and structured output.