Cursor Alternative Backend: Any Model, ₹ Billing

Cursor's built-in AI runs on a fixed model at a fixed price. unoblox lets you keep Cursor's editor and point it at ANY model instead — Claude, GPT-4o, DeepSeek, or the free Qwen3 1.7B — through one OpenAI-compatible endpoint billed in rupees on a monthly GST invoice.

Why swap Cursor's backend for unoblox

• Keep the editor: Cursor's UI, keyboard shortcuts, and Cmd+K/Cmd+L flows are unchanged
• Pick the model per task: cheap models for boilerplate, Claude Opus for architecture
• One ₹ invoice: every model you use rolls into a single monthly GST bill, ITC-eligible
• Works beyond Cursor: the same setup works in Windsurf, Continue.dev, and any editor that accepts a custom OpenAI-compatible endpoint

Setup in Cursor's settings

Open Cursor → Settings → Models → add a custom OpenAI-compatible provider:

API Provider: OpenAI API-compatible
Base URL: https://api.unoblox.ai/v1
API Key: ub-gw-YOUR_API_KEY
Model: openai/gpt-4o (or any unoblox model id)

Restart Cursor. Cmd+K, Cmd+L, and chat now route through unoblox; the rest of the editor behaves exactly as before.

Match the model to the task

Task	Model	₹ per 1M (in / out)
Boilerplate, comments, cleanup	Qwen3 1.7B	FREE
Tests, quick rewrites	DeepSeek V4 Flash	₹9.07 / ₹18.14
General coding, bug fixes	GPT-4o	₹252 / ₹1008
Architecture, hard reasoning	Claude Opus	₹504 / ₹2520

Change the model field per session — Cursor applies it immediately, no restart needed.

Illustrative cost split

A developer generating roughly 2 million tokens of editor output in a day, split by task difficulty, might land like this: 1.2M tokens on Qwen3 1.7B (free) for boilerplate, 700K on DeepSeek V4 Flash (₹18.14 per 1M output tokens) for routine code, and 100K on Claude Opus (₹2,520 per 1M output tokens) for the hard parts — roughly ₹0 + ₹12.70 + ₹252, or about ₹265 for the day, instead of paying premium rates for every keystroke.

Keyboard shortcuts (unchanged)

• Cmd+K / Ctrl+K: generate code inline
• Cmd+L / Ctrl+L: ask a question in chat
• Cmd+Shift+L / Ctrl+Shift+L: full-file refactor — swap to a cheaper model first, since it uses more tokens
• Cmd+I / Ctrl+I: inline edit from the context menu

Frequently asked questions

Will Cursor complain about a non-Anthropic backend? It may run a one-time connectivity check; that's normal. Cursor treats any OpenAI-compatible base URL as a valid provider, so there's no repeated warning after the first successful call.

Does every Cursor feature work with every model? Most do. A few advanced flows lean on stronger reasoning — if a cheap model struggles, switch that session to Claude Opus or GPT-4o and re-run.

Can I still use Cursor's original Claude backend? Yes. Switching back is a one-click settings change; nothing about your code or history is affected either way.

What happens if a request fails partway through? You're billed only for the tokens actually processed. Re-running a failed request starts a fresh, separately billed call.

How do I confirm the connection is working? Press Cmd+K, type a comment, and check for a completion. Your unoblox dashboard logs the request with model, tokens, and ₹ cost within seconds.

Can I use this setup with Windsurf or Continue.dev instead of Cursor? Yes — any editor with an OpenAI-compatible custom-provider option takes the same base URL and key.

Get started in rupees → https://unoblox.ai/sign-in