# Docs

Quick start

OrbitRouter uses the OpenAI Chat Completions format. Point any OpenAI SDK at the base URL below.

Base URL: /v1

API keys and credits

Every request needs an API key. Sign up to get $10 in free credits, then create a key on your dashboard. Send it as Authorization: Bearer sk-orbit-....

Each successful request is charged from your balance by input and output tokens, using the per-model prices on the models page (USD per 1M tokens). Failed requests are free. When your balance runs out, requests return 402.

Endpoints

MethodPathWhat it does
POST/v1/chat/completionsChat in OpenAI format. Set "stream": true to get SSE.
GET/v1/modelsList of models, including thinking and default_thinking for each.
GET/v1/statsPublic usage totals and leaderboards.
GET/healthHealth check.

Request fields

FieldNotes
modelModel id from /v1/models. Can end with a thinking suffix.
messagesThe roles system, user, assistant and tool. Images can be sent as image_url data URLs.
max_tokensMaximum length of the reply. If not set, there is no limit and the model stops on its own.
temperature, top_p, seed, stopPassed through to the model.
toolsOpenAI function-calling format.
response_format{"type": "json_object"} forces the reply to be JSON.

Thinking

HowExample
Model suffixgpt-oss-120b:high, qwen3.6-35b:think, gemma4-31b:nothink
Request field"think": true or false
OpenAI field"reasoning_effort": "low" | "medium" | "high" | "none"

Models that cannot think ignore these settings. Reasoning text is returned in message.reasoning_content.

Streaming

Errors

CodeMeaning
400Bad request body or bad parameter.
401Missing or invalid API key.
402Out of credits.
404Unknown model. Check /v1/models.
429Rate limited. Retry later.
502 / 504Model unavailable or timed out. Retry.