Chat API

One key,
access everylarge model

Compatible with OpenAI (Chat Completions / Responses) and the Anthropic protocol. Swap base_url for Lumen and call Claude, GPT, Gemini, DeepSeek and more from your existing SDK, client, Claude Code or Codex, billed by token.

Create a key in the console →

Endpoints

POST/v1/chat/completions

OpenAI Chat Completions compatible. Works with most clients and SDKs.

POST/v1/responses

OpenAI Responses compatible. Works with the Codex CLI and the latest OpenAI SDK.

POST/v1/messages

Anthropic Messages compatible. Works with Claude Code and the Anthropic SDK.

Base URL https://lumen.mengfanlab.top· Every endpoint supports streaming (SSE) and non-streaming, with response shapes matching the corresponding official protocol.

Authentication

Authenticate with a Bearer key

Create a key starting with sk-lumen- in the console and put it in the request header. Each call is billed from your account credits at the unit price of the chosen model and group.

Authorization: Bearer sk-lumen-•••

· The key is both identity and quota, so keep it safe and revoke it in the console if leaked.
· /v1/messages uses the Anthropic protocol; authenticate with Authorization: Bearer (not x-api-key).

Quick start

Start a conversation

Pick either protocol. The examples below use Claude Opus 4.8 and are all streaming requests.

POST /v1/chat/completions · OpenAI compatible

curl https://lumen.mengfanlab.top/v1/chat/completions \
  -H "Authorization: Bearer sk-lumen-•••" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-opus-4-8",
    "stream": true,
    "messages": [
      { "role": "user", "content": "用一句话介绍你自己" }
    ]
  }'

POST /v1/messages · Anthropic compatible

curl https://lumen.mengfanlab.top/v1/messages \
  -H "Authorization: Bearer sk-lumen-•••" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-opus-4-8",
    "max_tokens": 1024,
    "stream": true,
    "messages": [
      { "role": "user", "content": "用一句话介绍你自己" }
    ]
  }'

Request parameters

Common fields

Field semantics match the official docs of the chosen protocol; here are the most commonly used ones.

modelstringrequired

Model code, see "Supported models" below.

messagesarrayrequired

List of conversation messages, structured per the corresponding protocol (role + content).

streamboolean

Defaults to true for SSE streaming; set to false for a single response.

max_tokensinteger

Maximum tokens to generate. Required for Anthropic /v1/messages.

temperaturenumber

Sampling temperature, controls randomness.

Other fields (tools, tool_choice, system, stop, etc.) are passed through to the upstream per protocol with identical behavior.

Streaming

SSE streaming responses

When stream is true, the response is text/event-stream, returning incremental content chunk by chunk, with the same shape as the official APIs.

· OpenAI: each frame is a data: {...} chunk, the last frame carries usage, ending with data: [DONE].
· Anthropic: an event stream of message_start / content_block_delta / message_delta / message_stop.
· Billing is settled after the stream ends, based on actual token usage.

Supported models

Pass one of these codes

Just put the code in the model field. The full list and live unit prices are in the console.

claude-opus-4-8Claude Opus 4.8

claude-opus-4-7Claude Opus 4.7

claude-opus-4-6Claude Opus 4.6

claude-sonnet-4-6Claude Sonnet 4.6

gpt-5.5GPT-5.5

gemini-3.1-pro-previewGemini 3.1 Pro

gemini-3-pro-previewGemini 3 Pro

gemini-3-flash-previewGemini 3 Flash

deepseek-v4-proDeepSeek V4 Pro

deepseek-v4-flashDeepSeek V4 Flash

kimi-k2.6Kimi K2.6

glm-5GLM-5

minimax-m3MiniMax M3

qwen3.7-maxQwen3.7 Max

Connect Claude Code

Point Claude Code at Lumen

Set the environment variables, then run claude directly. base_url should not include /v1; Claude Code appends /v1/messages automatically.

ANTHROPIC_BASE_URL=https://lumen.mengfanlab.top \
ANTHROPIC_AUTH_TOKEN=sk-lumen-••• \
ANTHROPIC_MODEL=claude-opus-4-8 \
ANTHROPIC_SMALL_FAST_MODEL=claude-sonnet-4-6 \
claude

Authenticate with ANTHROPIC_AUTH_TOKEN (sends Authorization: Bearer); use Lumen model codes. Some low-price groups are restricted to the Claude Code client only.

Connect Codex

Point the Codex CLI at Lumen

Codex uses the OpenAI Responses protocol. After installing, write two files under ~/.codex: config.toml sets base_url and the model, auth.json holds the key.

1 · Install the Codex CLI

npm i -g @openai/codex@latest

2 · ~/.codex/config.toml

model_provider = "lumen"
model = "gpt-5.5"

[model_providers.lumen]
name = "lumen"
base_url = "https://lumen.mengfanlab.top/v1"
wire_api = "responses"
requires_openai_auth = true

3 · ~/.codex/auth.json

{
  "OPENAI_API_KEY": "sk-lumen-•••"
}

4 · Run

codex

· base_url must include /v1; Codex appends /v1/responses automatically; wire_api must be responses.
· Put the key in auth.json's OPENAI_API_KEY (starting with sk-lumen-); Codex sends it as Authorization: Bearer. Use a key from a regular group (cc_only groups are Claude Code only).
· Set model to a Lumen model code, see "Supported models" above; restart codex after changing the config.

Error codes

400

Invalid request body.

401

Key missing, invalid, or revoked.

402

Insufficient account credits.

403

This group is restricted (e.g. Claude Code client only).

404

The requested model does not exist in this group.

429 / 5xx

Upstream rate limit or error; the gateway passes the upstream status through unchanged.

Questions or partnerships, email [email protected].

One key,access everylarge model