Skip to main content

China Providers

💡TL;DR

Use the provider preset matching your account and deployment. Regional endpoint paths, model names and permissions are not interchangeable, even when providers expose the same OpenAI-compatible protocol. Verify a small task before choosing a provider for a workload.

Setup​

Create an API account with the chosen provider, obtain its key, and select the corresponding Notemd preset. Confirm the account's endpoint region and model entitlement. Run Test connection and a disposable note task; preset defaults do not guarantee current model availability.

Endpoint And Authentication​

PresetDefault modelDefault base URL
DeepSeekdeepseek-v4-prohttps://api.deepseek.com
Qwenqwen3-235b-a22bhttps://dashscope.aliyuncs.com/compatible-mode/v1
Qwen Codeqwen3-coder-plushttps://dashscope.aliyuncs.com/compatible-mode/v1
Doubaoep-xxxxxxxxxxxxxxxxhttps://ark.cn-beijing.volces.com/api/v3
Moonshotkimi-k2-0905-previewhttps://api.moonshot.cn/v1
Xiaomi MiMomimo-v2.5-prohttps://api.xiaomimimo.com/v1
GLMglm-5https://open.bigmodel.cn/api/paas/v4
Z AIglm-5https://api.z.ai/api/paas/v4
MiniMaxMiniMax-M2.7https://api.minimaxi.com/v1
Baidu Qianfanernie-4.5-turbo-32khttps://qianfan.baidubce.com/v2
SiliconFlowQwen/QwQ-32Bhttps://api.siliconflow.cn/v1
Huawei Cloud MaaSDeepSeek-V3https://api.modelarts-maas.com/v1

Replace Doubao's placeholder with an Ark endpoint ID or a model ID your deployment supports. GLM and Z AI have different regional presets. A key for one account region or subscription path may not work on another endpoint; use the provider's account instructions rather than choosing solely by your physical location.

Model Discovery​

Use model discovery where the provider permits it. Huawei Cloud MaaS uses deployment-aware discovery; other listed presets have OpenAI-compatible discovery paths where supported. Lack of listing permission does not necessarily mean an explicitly configured model cannot generate.

Connection-test behavior also differs by preset. DeepSeek can succeed from a model-list response alone, while several regional presets perform a short chat request. A real task remains necessary to verify its prompt and output requirements.

Troubleshooting​

  • Key accepted but generation denied: check model entitlement, deployment and account region.
  • Ark model rejected: confirm whether that endpoint expects a deployment ID or model ID.
  • Truncated reasoning response: inspect requested output limits and provider-specific thinking settings.
  • Network error: test the configured endpoint from the device running Obsidian and inspect the reported transport error.

When To Use​

Choose according to reachable account endpoints, data-handling requirements and measured task output. Region or model branding is not a quality, privacy or cost guarantee. Provider overview also lists gateways and local servers when those better match your deployment.

Next Steps​