Providers
All 24 supported AI model providers and their capabilities.
Provider architecture
Yunxin gives each provider an independent adapter that speaks the provider's native API. You always call Yunxin at https://api.yuhuanstudio.com/v1 with your sk- key — Yunxin then routes the request to the right upstream and translates between formats as needed.
- Three request formats, natively supported — send requests as OpenAI Chat Completions, the OpenAI Responses API, or the Anthropic Messages API. These are real implementations, not transpiled shims.
- Native routing — each provider is reached through its own best-suited API, so provider-specific features are preserved.
- Runtime-driven — providers and models are configured in the database; new ones can be added without a redeploy.
- Graceful fallback — if a provider is unavailable, requests can fall back to alternatives (surfaced via
X-Fallback-*response headers).
Addressing a provider
Route to a provider by prefixing the model id with the provider slug: provider/model (for example openai/gpt-4o, anthropic/claude-..., google/gemini-...). Auto-routing can also select a provider by category when you omit the prefix.
All providers
Capabilities are per-model. Use the Models API to discover what each provider currently offers.
| Provider | Slug | Strengths |
|---|---|---|
| Alibaba (Tongyi/Qwen) | alibaba | Multilingual, reasoning, vision, embeddings, image/video gen |
| Anthropic | anthropic | Claude, extended thinking, vision/PDF, prompt caching, server tools |
| Azure AI Foundry | azure | Enterprise/regional, OpenAI-compatible chat, embeddings, vision |
| Cloudflare (Workers AI) | cloudflare | Edge inference, chat, embeddings, image, STT |
| DeepSeek | deepseek | Cost-effective chat + reasoning, strong math/coding |
| GitHub Models | github | Developer-friendly, OpenAI-compatible chat, vision, embeddings |
| Google (Gemini) | google | Native multimodal, long context, thinking, image/video gen |
| Groq | groq | Ultra-fast LPU inference, chat, vision, STT/TTS |
| Hugging Face | huggingface | Largest model hub, Serverless Inference API |
| LM Studio | lmstudio | Desktop GUI, local inference |
| MiniMax | minimax | Chat, TTS, voice clone/design, video, music |
| Mistral AI | mistral | Open-weight models, code generation |
| ModelScope | modelscope | Chinese models, multimodal |
| Moonshot | moonshot | Long context, document analysis |
| NVIDIA NIM | nvidia | GPU-accelerated, enterprise-grade |
| Ollama | ollama | Simple local deployment |
| OpenAI | openai | Full suite: Chat, Responses, Realtime, vision, audio, embeddings |
| OpenRouter | openrouter | Gateway to many models, automatic fallback |
| TokenRouter | tokenrouter | Unified gateway, OpenAI-compatible |
| vLLM | vllm | High-throughput self-hosted serving |
| Volcengine | volcengine | ByteDance/Doubao ecosystem, video generation |
| xAI | xai | Grok models, reasoning, real-time knowledge |
| Z.AI | zai | GLM, multimodal, thinking, function calling |
| Zhipu AI | zhipu | GLM, strong bilingual support |
Yunxin exposes every provider through its three request formats — OpenAI Chat Completions, the OpenAI Responses API, and the Anthropic Messages API — and routes each request to the provider's native API. You write to one of these formats; Yunxin handles the rest.
Provider configuration
Providers are configured at the system level by administrators:
- Navigate to Dashboard → Admin → Providers
- Enable the desired provider
- Configure API credentials
- Set custom endpoint URLs if needed (for self-hosted instances)
Provider status
Public model and provider counts (no authentication required) are available at GET /v1/stats/public. It returns aggregate counts (total_models, total_providers, active_providers), an overall_status (operational, degraded, down, maintenance, partial_maintenance), uptime_sla, avg_latency_ms, total_requests_24h, and a per-service services_status list:
curl https://api.yuhuanstudio.com/v1/stats/publicTo list every provider with its enabled/health state, query GET /v1/providers (authenticated with a dashboard session). Each entry includes name, display_name, is_enabled, is_healthy (null when never checked), model_count, capabilities, and latency_ms:
curl https://api.yuhuanstudio.com/v1/providers \
-H "Authorization: Bearer $YUNXIN_SESSION_TOKEN"To inspect what is currently available to your API key, query the Models API:
curl "https://api.yuhuanstudio.com/v1/models?provider=openai" \
-H "Authorization: Bearer $YUNXIN_API_KEY"Related topics
- Authentication — how to create and use API keys
- Models API — discover models and capabilities
- Rate Limits — understand usage limits
- Vision & Documents — multimodal capabilities
- Function Calling — tool use guide
How is this guide?