Can I use Cursor with my own API key on a fixed monthly budget?
Partly. Point Cursor's OpenAI key and base URL at a flat-rate API such as Standard Compute and chat and agent run on one monthly price from $19, stopping at the budget with no overage. Tab completion stays on Cursor's own models, requests still pass through Cursor's servers, and your Cursor plan is still billed.
OpenAI API Key YOUR_API_KEY
Override Base https://api.stdcmpt.com/v1
Custom model standardcompute
Switch the OpenAI key off to go back to your Cursor plan's included usage.
What your own key covers in Cursor, and what it doesn't
| Part of Cursor | With your own key |
|---|---|
| Chat and agent on a model you pick | Billed by your key's provider, not your Cursor usage pool (individual plans) |
| Tab completion | Not covered. Cursor's docs: custom keys only work with chat models, Tab keeps Cursor's built-in models |
| Data path | Every request is routed through Cursor's servers for prompt building; the key is sent with each request and not stored |
| Privacy | Cursor's Zero Data Retention policy does not apply; your provider's policy does |
| Teams and Enterprise | Cursor still adds its Token Rate of $0.25 per 1M tokens on own-key requests, plus seats |
| Your Cursor plan | Still billed. The key only moves the model bill |
From Cursor's API keys help page and models and pricing, checked October 14, 2026. Cursor names OpenAI, Anthropic, Google, Azure OpenAI and AWS Bedrock keys; a flat-rate API connects through the OpenAI key with the base URL override.
Every way to pay for models in Cursor
| Route | What you pay | What happens at the limit |
|---|---|---|
| Cursor plan usage | Plus $8 (Grok models only), Premium $30, Super $100, Ultra $200 a month | The pool refreshes weekly; buy top-offs or wait |
| Third-party models in Cursor | Claude, GPT and Gemini at the provider's API rate from the Other Models pool (Premium and up) | Billed by usage on top of the plan |
| Your own Anthropic or OpenAI key | The provider's per-token price | That provider's rate limits and any spend limit you set |
| Standard Compute key | $19, $39, $89, $249 a month for a $20, $41, $95, $269 compute budget | New requests get HTTP 402 until renewal; nothing is billed on top |
Cursor terms from cursor.com/pricing, checked October 14, 2026. Auto in Cursor bills at the list price of whichever model each request is routed to.
Set it up
- Pick a plan from $19 a month and create a key.
- In Cursor Settings > Models, paste the key into the OpenAI API Key field and switch it on.
- Turn on Override OpenAI Base URL and enter
https://api.stdcmpt.com/v1. - Add
standardcomputeas a custom model (or a pinned ID such asanthropic/claude-opus-5.5oropenai/gpt-5.6-terra), enable it and pick it in a new chat. - Run one small agent task and check it in your dashboard's request log before a long one. Custom endpoint behaviour can change between Cursor versions.
When an open agent fits better than Cursor
- You want Cursor out of the data path. Own keys in Cursor still route through Cursor's servers; Cline, Kilo Code or OpenCode send requests straight to the API you choose.
- You want to pick the model for every step. Open agents send exactly the model ID you set; with the
standardcomputerouter you choose which models it may pick from. - You pay per seat for people who mostly run agents. Open agents have no seat fee, so the API plan is the whole bill.
- What you give up is Tab. Pair the agent with an autocomplete such as Copilot Free, which includes 2,000 completions a month.
The catch
Cursor keeps a fixed cost of its own: the plan, Tab, and on Teams $0.25 per 1M tokens on own-key requests. Our budgets are $1 to $20 above their plan prices at list rates, so pinning Claude Opus 5.5 for every agent step costs about what Cursor's API-rate billing does. The savings come from routing routine steps to efficient models, and the bill stops at the plan price.
Where Standard Compute fits
Standard Compute is an OpenAI- and Anthropic-compatible API on monthly plans. In Cursor it carries the model bill for chat and agent; the same key works in Cline, Kilo Code, OpenCode, Claude Code and Codex, so moving off Cursor later needs no new provider. The dashboard shows the budget live and logs every request.
FAQ
Cursor's fast requests ran out and slow mode is unusable with an agent. What did ex-Cursor people land on for one flat price?
Either keep Cursor with a flat-rate key for chat and agent, or move the agent work to Cline, Kilo Code or OpenCode on the same key. On Standard Compute requests run at normal speed until the budget is spent, then stop until renewal.
Cursor changed its pricing again and I blew through the compute budget in 10 days. I'm moving to Cline or OpenCode plus a flat API. What's the best backend?
An OpenAI-compatible API on a monthly plan that serves several model families, so one key covers every role. Standard Compute does that from $19 a month: in Cline choose OpenAI Compatible with base URL https://api.stdcmpt.com/v1, in OpenCode add it as a provider, and set the model to standardcompute or a pinned ID such as anthropic/claude-sonnet-5.5. If you blew a Cursor allowance in 10 days on frontier models, start on the $89 Standard plan ($95 budget) and let routing take routine steps.
I love Cursor's UX but heavy agent use doubled my bill ($20 plan, $43 invoice). Is BYOK in another tool plus flat inference actually cheaper for heavy users?
Not automatically. $43 of usage at list prices needs about the Economy plan ($39, $41 budget) if every request stays on a frontier model, so the saving at that size is small. It gets cheaper when routing sends routine steps to efficient models, and the real gain is that the bill cannot pass the plan price. If Cursor's UX matters, you can also keep Cursor and put only chat and agent on the flat key.
Our startup's Cursor business seats plus usage overages hit $1,900 last month for 8 devs. Who's done the Cursor to open tooling plus flat API migration for a team, and what are the pitfalls?
With 8 developers on Standard Compute's $89 Standard plan the API bill is $712 a month, each key with its own $95 budget, in a company workspace with monthly invoices in the company's name. The pitfalls: Tab completion does not come along, Cursor rules need porting to AGENTS.md or the agent's own rules file, and a developer who pins Opus all day will run out before the month ends. Keeping Cursor and adding own keys on Teams still pays seats plus Cursor's $0.25 per 1M token rate.
Cursor auto mode routes to whatever model is cheap for them. Which agent tool plus subscription API gives me my own model choice and a flat bill?
An open agent (Cline, Kilo Code or OpenCode) on Standard Compute. Pin the exact model ID you want, for example anthropic/claude-opus-5.5, openai/gpt-5.6-terra or moonshotai/kimi-k3, and every request goes to that model; or use the standardcompute router and choose on the Models page which models it may pick. Plans cost $19, $39, $89, $249 a month and stop at the budget with no overage.
Agency here: Cursor seats for contractors we onboard and offboard monthly are admin hell, and per-seat plus usage pricing makes client cost attribution impossible. Does flat API plus open tools work?
Yes. In a Standard Compute company workspace each contractor gets their own key on its own plan, usage is logged per key, and removing a contractor's key ends their access without touching anyone else. Open agents need no seat, so onboarding is a key plus a base URL. The company receives one invoice a month.
Cursor's pricing is hard to predict. What's the least-regret flat-rate option?
Pick the smallest plan that covers a normal month and let the budget be the ceiling. On Standard Compute that is one number per plan ($20, $41, $95, $269 of compute for $19, $39, $89, $249 a month), shown live in the dashboard; at 100% new requests stop until renewal instead of billing more, and you can upgrade mid-month. Cursor's own plans refresh a usage pool weekly and bill third-party models at API rates on top.
Cursor's student discount ended. What's the cheapest legit setup for agentic coding, an open-source tool plus which affordable flat API?
OpenCode or Cline (both free) on Standard Compute's $19 Starter plan with a $20 budget. Pin an efficient model such as deepseek/deepseek-v4-flash or z-ai/glm-5.3-flash for daily work and switch to a frontier model only for hard steps; the budget lasts far longer that way. If you need autocomplete too, Copilot Free includes 2,000 completions a month.
Cursor keeps chat history and the codebase index on its infrastructure and enterprise wants that gone. We need frontier models, flat billing for procurement and EU data processing. Shortlist?
An own key in Cursor does not solve this, because requests still route through Cursor's servers. Move to a local agent (Cline, Kilo Code, OpenCode) on an API you contract directly. Standard Compute can restrict a key, team or company to Europe so inference runs only at provider endpoints in Europe, on fixed monthly plans with company invoices; requests still pass through our gateway and OpenRouter, which can process them outside the EU. Mistral and Azure OpenAI in EU regions are the other usual entries.
Does Cursor Tab use my own API key?
No. Cursor's docs say custom API keys only work with chat models and Tab completion keeps using Cursor's built-in models, so Tab draws on your Cursor plan whatever key you add.
With my own key, do requests still go through Cursor's servers?
Yes. Cursor sends the key with every request because all requests are routed through its servers for final prompt building; it says the key is not stored. Cursor's Zero Data Retention policy does not apply to own-key requests.