Configure
Provider endpoints reference
Every endpoint group in the built in profile catalog: plans, regions, dialects, base URLs, and current model sets — plus the credential semantics that distinguish look alike endpoints. The one line per provider overvi...
Documentation notice: these docs are currently generated with Codex and are being actively reviewed, expanded, and refined.
Every endpoint group in the built-in profile catalog: plans, regions, dialects, base URLs, and current model sets — plus the credential semantics that distinguish look-alike endpoints. The one-line-per-provider overview is in the supported matrix; how to use profiles is the model profiles guide.
Two rules that apply across providers:
- A plan is a credential, not just a URL. Several providers serve pay-as-you-go API and subscription plans (Coding Plan, Token Plan, Agent Plan) on the same or similar base URLs — the difference is which kind of key is accepted. Pick the endpoint group matching the credential you actually bought.
- Model lists are curated. The catalog carries current flagship/coding models per endpoint, not the provider's full marketplace. Anything missing works via a custom profile with the same base URL.
| Endpoint group | Dialects | Base URL | Key models |
|---|
moonshot-global | openai-chat, anthropic | https://api.moonshot.ai/v1 · /anthropic | kimi-k2.7-code, kimi-k2.7-code-highspeed, kimi-k2.6, kimi-k2.5, moonshot-v1-* |
moonshot-cn | openai-chat, anthropic | https://api.moonshot.cn/v1 · /anthropic | same family |
kimi-coding | openai-chat, anthropic | https://api.kimi.com/coding/v1 · /coding/ | kimi-for-coding |
- Kimi Coding is a separate product with its own key and a single documented model — not a region of the Moonshot API.
- K2.x models are 256K-context and multimodal; K2.7 Code always uses thinking.
| Endpoint group | Dialects | Base URL | Key models |
|---|
api-cn-beijing | openai-chat, openai-responses, anthropic | https://dashscope.aliyuncs.com/compatible-mode/v1 · /apps/anthropic | qwen3.7-max/plus, qwen3.6-flash, deepseek-v4-pro, kimi-k2.6, glm-5.2, MiniMax-M2.5 |
coding-plan (intl) | openai-chat, anthropic | https://coding-intl.dashscope.aliyuncs.com/v1 · /apps/anthropic | qwen3.7-plus, qwen3.6-plus, kimi-k2.5, glm-5, MiniMax-M2.5 |
coding-plan-cn | openai-chat, anthropic | https://coding.dashscope.aliyuncs.com/v1 · /apps/anthropic | same as intl |
token-plan-cn | openai-chat, openai-responses, anthropic | https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1 · /apps/anthropic | broad flagship set incl. third-party models |
- Pay-as-you-go, Coding Plan, and Token Plan are three different credentials; Coding/Token Plan keys do not work on the plain API endpoint and vice versa.
- The Responses dialect is limited to the tool-capable Qwen subset (
qwen3.7-max/plus, qwen3.6-plus/flash) — built-in search/code/web tools require it.
- Bailian's international regional endpoints (US / Singapore / EU workspace domains) are documented by Alibaba but not in the catalog — use a custom profile if you need them.
| Endpoint group | Dialects | Base URL | Key models |
|---|
| default | openai-chat, anthropic | https://api.deepseek.com · /anthropic | deepseek-v4-pro, deepseek-v4-flash |
- Note the OpenAI base has no
/v1 suffix.
- The Anthropic-compatible endpoint maps Claude tiers officially: Opus-class requests →
v4-pro, Sonnet/Haiku-class → v4-flash. Image/document content blocks are not supported there.
| Endpoint group | Dialects | Base URL | Key models |
|---|
api-global | openai-chat, openai-responses, anthropic | https://api.minimax.io/v1 · /anthropic | MiniMax-M3, M2.7(-highspeed), M2.5(-highspeed) |
api-cn | openai-chat, anthropic | https://api.minimaxi.com/v1 · /anthropic | same family |
token-plan-global | openai-chat, openai-responses, anthropic | same base as api-global | same family |
token-plan-cn | openai-chat, anthropic | same base as api-cn | same family |
- API and Token Plan share base URLs; the split exists because pay-as-you-go API keys and Token Plan Subscription Keys are different credentials.
MiniMax-M3 is the flagship: 1M context, image/video input, thinking; the Responses dialect adds reasoning controls.
| Endpoint group | Dialects | Base URL | Key models |
|---|
pay-as-you-go | openai-chat, anthropic | https://api.xiaomimimo.com/v1 · /anthropic | mimo-v2.5-pro, mimo-v2.5, mimo-v2-pro, mimo-v2-omni, mimo-v2-flash |
token-plan-cn | openai-chat, anthropic | https://token-plan-cn.xiaomimimo.com/v1 · /anthropic | same family |
mimo-v2.5-pro is the recommended coding default (1M context, 128K output, web search, function calls); v2-omni adds full-modal input.- Token Plan Singapore/Amsterdam regional endpoints exist upstream but are not in the catalog — custom profile if needed.
| Endpoint group | Dialects | Base URL | Key models |
|---|
ark-api | openai-chat, openai-responses, anthropic | https://ark.cn-beijing.volces.com/api/v3 · /api/compatible | versioned deployments: doubao-seed-2-0-code-preview-260215, …-pro-260215, …-lite-260428, deepseek-v4-pro-260425 |
coding-plan | openai-chat, anthropic | …/api/coding/v3 · /api/coding | plan aliases: ark-code-latest, doubao-seed-2.0-code/pro, minimax-m3, glm-5.2, deepseek-v4-pro, kimi-k2.6 |
agent-plan | openai-chat, anthropic | …/api/plan/v3 · /api/plan | same alias set |
- Ark API uses versioned deployment ids; the plans use rolling aliases (
ark-code-latest) — the two id styles are not interchangeable across endpoint groups.
- Plans expose flagship third-party models (MiniMax, GLM, DeepSeek, Kimi) under one Volcengine credential.
| Endpoint group | Dialects | Base URL | Key models |
|---|
global | openai-chat | https://api.z.ai/api/paas/v4 | glm-5.2, glm-5-turbo, glm-5v-turbo, glm-5.1, glm-4.7, glm-4.5-air |
cn | openai-chat | https://open.bigmodel.cn/api/paas/v4 | same family |
coding-global | openai-chat, anthropic | …/api/coding/paas/v4 · https://api.z.ai/api/anthropic | glm-5.2, glm-5-turbo, glm-4.7, glm-4.5-air |
coding-cn | openai-chat, anthropic | …/api/coding/paas/v4 · https://open.bigmodel.cn/api/anthropic | same set |
- The Anthropic-compatible endpoints belong to the Coding Plan product, not the general API — a general API key will not authenticate there.
| Endpoint group | Dialects | Base URL | Key models |
|---|
gemini-api | gemini, openai-chat | https://generativelanguage.googleapis.com (+/v1beta/openai) | gemini-3.1-pro-preview, gemini-3-flash-preview, gemini-2.5-pro/flash(-lite), gemini-3.1-flash-lite |
google-accounts | gemini | https://cloudcode-pa.googleapis.com | same family |
vertex-openai-compatible | openai-chat | user-supplied | google/…-prefixed ids |
- Three auth shapes:
gemini-api = API key; google-accounts = Google OAuth (code-assist flow, handled by a dedicated adapter); Vertex = your own endpoint URL with google/-prefixed model ids.
| Endpoint group | Dialects | Base URL | Key models |
|---|
| default | openai-chat, openai-responses | https://api.x.ai/v1 | grok-4.3, grok-build-0.1 |
| Endpoint group | Dialects | Base URL | Key models |
|---|
| default | openai-chat | https://integrate.api.nvidia.com/v1 | nvidia/nemotron-3-super-120b-a12b, nvidia/nemotron-3-nano-30b-a3b, qwen/qwen3-coder-480b, openai/gpt-oss-120b, moonshotai/kimi-k2.6 |
| Endpoint group | Dialects | Base URL | Key models |
|---|
| default | openai-chat | https://openrouter.ai/api/v1 | anthropic/claude-sonnet-4.5, openai/gpt-4o, deepseek/deepseek-chat, google/gemini-2.5-pro |
- Catalog carries four representative ids; OpenRouter's whole marketplace works — type any model id the router accepts.
| Endpoint group | Dialects | Base URL | Key models |
|---|
| default | openai-responses | user-supplied resource URL | user-supplied deployment name |
- Azure endpoints are per-resource (
https://{resource}.openai.azure.com/openai/v1/) and models are your deployment names — the catalog intentionally ships both fields empty.
Provenance. Base URLs and model sets mirror src/resources/profile-catalog/ (the shipped catalog); plan/credential semantics were verified against official provider documentation in the 2026-06-18 audits (reports/provider-plan-api-audit-2026-06-18.md, reports/provider-profile-official-verification-2026-06-18.md). The catalog deliberately has no model-discovery metadata — model lists are curated, and a custom profile covers anything beyond them.
Source anchors: src/resources/profile-catalog/*.json (endpoints, base URLs, models), src/core/src/profiles/catalog.rs (loader), reports/provider-plan-api-audit-2026-06-18.md + reports/provider-profile-official-verification-2026-06-18.md (official-doc verification).
Last verified: v0.7.11
