How to Use Cursor with an Affordable AI API

As of 2026-10-10, Claude Sonnet 4.6 costs $1.15 / $5.74 per million input / output tokens this way, against $3.00 / $15.00 at Anthropic's list price. The same key also works in VS Code with Cline or Continue and in JetBrains plugins that accept an OpenAI-compatible endpoint.

Your own key tends to be cheaper when you use Cursor lightly or unevenly, and it lets you pick exactly which model answers. A subscription can make more sense if you code all day and stay within the plan's limits.

One caveat: some Cursor features may only run on Cursor's own models and not on a custom key. In past versions this has included things like Tab autocomplete and some agent or apply features. Which features use your key can change between releases, so check the API keys section of the Cursor docs for your version. For a broader comparison, see Claude Code vs Cursor: cost comparison.

1. Sign up at apivai.com with your email and a verification code
2. Top up your balance: from $10 by card or crypto, or ¥20 via Alipay / WeChat Pay
3. Create a key on the Dashboard, optionally with its own budget, and copy it

The setting names below are the ones in current versions; they may be worded slightly differently in yours.

1. Open Cursor Settings (Cmd+, on macOS, Ctrl+, on Windows and Linux) and go to the Models section
2. Find the setting named roughly Override OpenAI Base URL, turn it on and enter:

1. Paste your APIVAI key into the OpenAI API Key field
2. Click Verify (or the equivalent button). Turn the base URL override on before you verify, otherwise Cursor checks the key against OpenAI and rejects it.

In the same Models section, add each model ID you want as a custom model, for example claude-sonnet-4-6, then enable it. Type the ID exactly as shown in the table below. To check which IDs your key can see, request GET https://api.apivai.com/v1/models with the key.

Prices per 1M tokens as of 2026-10-10. Every model and the current price list are on the pricing page.

• Questions, small edits, renames, explaining code: Claude Haiku 4.5 or GPT-5.6 Luna. They are fast and cost little per turn.
• Everyday feature work and bug fixes: Claude Sonnet 5.5 (or the older Claude Sonnet 4.6). This is the sensible default for most people.
• Large refactors, tricky bugs across many files, design decisions: Claude Opus 5.5.
• Problems where Opus gets stuck: Claude Fable 5.1, the most capable and most expensive option. Use it for single hard tasks rather than as a default.
• A second opinion: switch to GPT-5.5 when you want a different model family to review the same code.

## How much does Cursor cost with your own key?

Costs depend on how many tokens each request sends. Three typical cases with Claude Sonnet 4.6:

• One chat turn of about 4,000 input and 1,500 output tokens: $0.0132 ($0.0345 at list price).
• One agent run that reads many files and makes 15 requests of about 40,000 input and 2,000 output tokens each: $0.862 ($2.25 at list price). Agent runs cost more because each step resends the files and conversation so far.
• A month of 30 chat turns a day for 30 days: $11.89 ($31.05 at list price). Add two agent runs a workday (about 44 a month) and that adds roughly $37.94.

The same month of chat turns on Claude Haiku 4.5 would be $3.96. The Dashboard shows the tokens and cost of every request, so you can see your real numbers after a day or two. The Claude API pricing comparison has the full price list next to official prices.

Cline is a VS Code extension with an agent mode. In current versions you open Cline's settings and choose an API Provider. There are two ways to connect:

• OpenAI Compatible provider: set the Base URL to https://api.apivai.com/v1, paste your APIVAI key as the API key, and enter a model ID such as claude-sonnet-4-6 or gpt-5.5.
• Anthropic provider: paste your key and enable the option named roughly Use custom base URL, then enter https://api.apivai.com (no /v1, because Cline adds the path itself). Pick a Claude model from the list.

Either works. The Anthropic provider speaks the native Claude format; the OpenAI Compatible one lets you mix Claude and GPT models with one setting.

Continue is configured in a file. Recent versions use config.yaml (older ones used config.json with the same fields). Add a model entry with provider openai and the APIVAI base URL:

Add one entry per model; Continue's docs list the other fields.

JetBrains IDEs (IntelliJ IDEA, PyCharm, WebStorm and others) work with any plugin that accepts an OpenAI-compatible endpoint. Continue has a JetBrains version that uses the same config file. Recent versions of JetBrains AI Assistant can also connect to third-party OpenAI-compatible providers; see the AI Assistant docs for the current menu. In any such plugin, the values are the same: base URL https://api.apivai.com/v1, your APIVAI key, and a model ID from the table. See OpenAI-compatible API for the general details.

• "Invalid API key" or 401: turn the base URL override on before you verify the key. Also check that the key was copied in full, has not been deleted, and has budget left on the Dashboard.
• 404: the base URL is wrong. OpenAI-format tools (Cursor, Cline's OpenAI Compatible provider, Continue) need /v1 exactly once. Anthropic-format settings (Cline's Anthropic provider) must not include /v1.
• Model not found: the model ID must match exactly, for example claude-sonnet-4-6, not a display name.
• 429: the default limit is 60 requests per minute per key. Agent runs can hit it; wait a moment, or ask support for a higher limit.

All API error codes and their fixes are collected in Claude API error codes.

Cursor's terms for using your own key have changed over time, so check Cursor pricing for what your plan allows. The API side needs no subscription: you only top up an APIVAI balance.

APIVAI does not store or inspect the content of requests or responses. Only usage metadata such as model, token counts, cost and time is kept for billing.

Yes. One key works with every model on the pricing page, in both the OpenAI format (https://api.apivai.com/v1) and the Anthropic format (https://api.apivai.com).

Possibly not. Some Cursor features may only work with Cursor's own models. In Continue you can set an autocomplete model yourself.

Give the editor its own key with a budget, use a cheaper model by default, and check the Dashboard, which lists every request with its cost. For terminal workflows, see the Claude Code setup guide.

Create an APIVAI account and top up from $10 to start.