Cursor already supports custom OpenAI-compatible endpoints. Qubax is one — 399 models behind one API, currently averaging ~78% below standard retail. Here's the 5-minute setup.
What you need
- A free Qubax API key (create one here — no credit card needed)
- Some credits (crypto or card, from $1)
- Cursor (any recent version)
Step 1: Get your Qubax key
Sign up at qubax.ai, go to API Keys, and create a key. It starts with qbx-.
Step 2: Point Cursor at Qubax
Open Cursor Settings → Models (⌘, on Mac). Under OpenAI API Key:
- API Key: paste your
qbx-key - Override Base URL:
https://api.qubax.ai/v1
Toggle Override OpenAI Base URL on, click Verify.
Step 3: Add the models you want
Click + Add model and add model IDs like:
claude-opus-4-8— frontier reasoning, $2.20/$11.02 per 1M (55% below retail)gpt-5-6-sol— fast frontier, $0.04/$0.20 per 1Mkimi-k3— the agentic coding value king, $0.03/$0.15 per 1Mglm-5-2— $0.02/$0.08 per 1M, shockingly good at code
Exact model IDs are listed for every model on our models page — search, copy the ID, paste into Cursor.
Step 4: Use it
Open the composer (⌘I), select your model from the dropdown, and code. Tab-complete, agent mode, and chat all route through Qubax.
What it costs
A realistic day of heavy agentic coding (~10M input / 2M output tokens) on Kimi K3:
- Standard retail: ~$1.00+/day
- Qubax: ~$0.24/day
Same models, same tokens — the pricing table with raw numbers is here.
Troubleshooting
- "Model not found" — check the exact ID on our models page; IDs match upstream (e.g.
deepseek-v4-flash, notdeepseek). - Verification fails — make sure the base URL ends with
/v1. - Streaming issues in agent mode — set your client to allow streaming; all Qubax models support SSE.
That's it. One base URL swap, 399 models, retail-beating prices. Get your free API key.
