Skip to main content
Corvex is a serverless, OpenAI- and Anthropic-compatible inference API. Your entire surface is simple: send inference requests and read your own usage.

​
Endpoints

Chat completions

POST /v1/chat/completions — OpenAI-compatible chat, streaming and non-streaming. This is the primary endpoint.

Text completions

POST /v1/completions — legacy text completions, served only by text-completion models. Other models return an error pointing you to /v1/chat/completions.

Responses

POST /v1/responses — OpenAI Responses API, adapted onto chat completions.

Embeddings

POST /v1/embeddings — OpenAI-compatible embeddings endpoint. No embedding model is currently deployed.

Messages (Anthropic)

POST /v1/messages — Anthropic-compatible Messages API for the Anthropic SDK and Claude Code, plus POST /v1/messages/count_tokens for token counts.

Models

GET /v1/models (alias /openai/v1/models) — list available models with per-model pricing and documentation links.

Usage

GET /api/v1/usage — read-only usage for your key, filterable by range, start, and end.

​
Authentication

All requests use a single virtual key with the sk-corvex- prefix. Send it as a Bearer token:
On the Anthropic /v1/messages surface, you may instead pass it as an x-api-key header. Legacy sk-bf- keys are still accepted.
Keys are created and managed only from the dashboard at tokenfactory.corvex.cloud. There is no API for creating or rotating keys.

​
Base URL

​
Models

Models are addressed by their bare Hugging Face repo id (namespace/model). Call GET /v1/models for the current catalog, pricing, and documentation links — it is the authoritative source. Currently available:
  • zai-org/GLM-5.2-FP8
  • moonshotai/Kimi-K2.7-Code
Already using the OpenAI or Anthropic SDKs? Point the base URL at https://api.tokenfactory.corvex.cloud and use your sk-corvex- key — most applications work without further changes.