Cursor with a discounted custom OpenAI endpoint

Cursor's credit pool runs out faster than you think

Cursor's $20/month Pro plan includes a credit pool that depletes based on which model auto-picked and how complex the task was. Heavy Composer + Agent users routinely burn through monthly credits in two weeks, hitting overage charges or rate limits exactly when productivity should ramp. Cursor exposes a 'Custom OpenAI endpoint' setting precisely for this scenario — bring your own backend, pay metered rates, no surprise caps.

Configure Cursor's custom OpenAI endpoint to point at Anvat

In Cursor → Settings → Models → OpenAI API Key, paste your Anvat key. In the same panel, set the Base URL to https://api.anvat.app/v1. Cursor now routes every AI call through Anvat at 30% off list. The credit pool stops mattering — you pay per token, see exact spend in the Anvat dashboard, and get access to Claude + Gemini + DeepSeek alongside OpenAI through the same key.

Why this beats the obvious alternative

• 30% off every model Cursor calls

  GPT-5 drops to $7/$21 effective, Sonnet 4.6 to $2.10/$10.50, Opus 4.8 to $10.50/$52.50.

• Stop hitting credit pool ceilings

  No artificial monthly cap — you pay metered, scale linearly with your work.

• Multi-provider through one key

  Switch between OpenAI, Anthropic, Google, DeepSeek in Cursor's model picker without juggling multiple keys.

• Same Cursor UX

  Composer, Agent mode, inline chat, autocomplete — all keep working exactly as before.

• Per-request cost visibility

  Anvat's dashboard logs every call with its actual discounted cost — useful for chargeback or personal tracking.

Quickstart

Cursor → Settings → Models

OpenAI API Key:    sk-anvat-...
Base URL:          https://api.anvat.app/v1

# Optional model overrides (set these in Custom Models panel)
gpt-5                  → claude-sonnet-4-6   # cheaper default
gpt-5-mini             → claude-haiku-4-5    # cheapest fast tier
o1                     → claude-opus-4-8     # deep reasoning

FAQ

Does Cursor's Composer / Agent mode work through Anvat?

Yes. Cursor calls the same /v1/chat/completions endpoint for all features. Anvat forwards every request unchanged — Composer's multi-file edits, Agent mode's autonomous loops, inline autocomplete all behave identically.

Will tab autocomplete still be fast?

Anvat adds <10ms gateway latency on top of provider response time. Cursor's tab completion (which uses small specialized models) is unaffected — Anvat is for the chat/composer/agent surface where you've configured a custom endpoint.

Can I keep Cursor's built-in Pro plan AND use Anvat?

Yes. Cursor's built-in models use Cursor's pool; your custom endpoint calls use Anvat's pool. They're independent — useful for hedging or A/B testing model quality between vendors.

What models can I access through Anvat in Cursor?

All 28+ frontier models in the Anvat catalog: every Claude tier (Opus, Sonnet, Haiku), GPT-5 family, Gemini 2.5, DeepSeek V4, Llama 4. Configure aliases in Cursor's Custom Models panel to remap names if needed.