Integration · Anysphere

Use Crazyrouter as a custom model provider in Cursor

Cursor's BYOK mode accepts any OpenAI-compatible endpoint. Enter the Crazyrouter base URL once and route chat and agent traffic through your own key with live per-token pricing.

Get an API key

Manual configuration

No scripts needed — three fields in the Models settings page. Verify must succeed before Cursor lets you pick the model; the most common reason it fails is a base URL with the wrong path.

Cursor → Settings → Models

text

OpenAI API Key:            sk-your-crazyrouter-key
Override OpenAI Base URL:  https://api.crazyrouter.com/v1
Model (add custom):        claude-opus-4-8   (or gpt-5.5, deepseek-v4-pro …)
→ click Verify

Custom keys only apply to chat/agent models. Tab completion keeps using Cursor's own models — that is a Cursor limitation.

Base URL rule

https://api.crazyrouter.com/v1 — This tool is OpenAI-compatible and appends /chat/completions (or /responses) itself. Use the /v1 base URL exactly as shown.

Recommended models

Live Crazyrouter prices, refreshed hourly. Any model in the catalogue works; these are the ones we'd start with.

Claude Opus 4.8

claude-opus-4-8

$3.25 in · $16.25 out /1M tokens

GPT-5.5

gpt-5.5

$3.25 in · $19.50 out /1M tokens

DeepSeek V4 Pro

deepseek-v4-pro

$0.67 in · $2.00 out /1M tokens

FAQ

Which base URL goes in 'Override OpenAI Base URL'?

Exactly https://api.crazyrouter.com/v1. Not the root domain, not a full /chat/completions path.

Why does Tab completion not use my key?

Cursor documents that custom API keys apply only to standard chat models; Tab and some agent features use Cursor's own models.

Which model should I add first?

claude-opus-4-8 or gpt-5.5 for agent mode; deepseek-v4-pro when you want a cheaper default for chat.

Costs jumped after switching

Give Cursor its own token with a quota cap and a narrow model allowlist, so one shared key isn't drained by agent loops.