Cursor publishes a fairly clean three-tier ladder on cursor.com/pricing: a Free plan, Pro at $20/month, and Business at $40/month per user. The plan numbers are simple. The cost reality isn’t, because Cursor’s billable unit — the “premium request” — is metered differently depending on which model you pick, and the moment you flip to BYO-key for Claude or GPT, you’re back on per-token API rates from Anthropic or OpenAI on top of your Cursor seat. This guide breaks every layer.

Methodology. Plan caps and per-request weights below come from cursor.com/pricing and the in-app usage dashboard, last reconciled in May 2026. Anthropic and OpenAI API rates are from anthropic.com/pricing and openai.com/api/pricing. Cursor adjusts request weights periodically — treat these numbers as accurate within a small margin, not gospel.

Tier ladder at a glance

Free (Hobby)
$0/month
  • Roughly 50 slow premium requests per month and around 200 fast premium requests as a one-time onboarding allowance
  • Tab autocomplete (cursor-small) included with generous limits
  • No privacy mode
  • Hard ceiling once the slow allowance runs out — you wait for the queue or upgrade
Business
$40/month/user
  • Everything in Pro
  • Privacy mode enforced — code never used for training
  • Centralized team billing and seat management
  • Admin dashboard, SSO, usage analytics
  • Same 500 fast premium request quota per user

The price spread between Free and Pro is meaningful. The price spread between Pro and Business is mostly about who you are, not how much you build — Business buys you privacy guarantees, SSO, and an admin pane. A solo founder with no co-founders and no enterprise customers staring at their procurement form has no reason to pay $40.

What a “premium request” actually means

This is where the pricing page stops being self-explanatory. A premium request in Cursor isn’t a flat unit. The same prompt costs you a different number of requests depending on which model you point Cursor at:

  • Claude Sonnet — counted as 1 request in most modes. The standard workhorse model. If you only ever use Sonnet, your 500-request Pro budget behaves like 500 prompts.
  • Claude Opus — weighted significantly higher per call. A single Opus request can consume the equivalent of multiple Sonnet requests. Heavy reasoning on Opus burns the budget five to ten times faster.
  • GPT-4.1 / GPT-class frontier models — weighted between Sonnet and Opus depending on context size and tool calls.
  • Auto / cursor-small — effectively free, no premium-request consumption.

So “500 fast premium requests” isn’t a single budget. It’s 500 if you stick to Sonnet; it’s 50–100 effective prompts if you live in Opus. Cursor publishes the multipliers on the pricing page and shows real-time consumption in the in-app usage panel — it’s worth opening that panel once a week to calibrate.

The corollary: model selection is now a budget decision, not just a quality decision. For routine refactors, sticking to Sonnet is the right call — it’s nearly always good enough and it preserves your premium budget. Reserve Opus for the genuinely hard problems where you’ve already failed twice with Sonnet. We expanded on this trade-off in our Cursor review and the three-way comparison with Lovable and Claude.

The hidden line item: BYO-key API costs

Cursor lets you bring your own Anthropic or OpenAI key. When you do, Cursor stops counting those calls against your premium-request budget — and starts charging the model vendor directly through your key. People assume this is cheaper. Sometimes it isn’t.

Current rates from anthropic.com/pricing and openai.com/api/pricing as of May 2026:

ModelInput / 1M tokensOutput / 1M tokens
Claude Sonnet 4.x$3.00$15.00
Claude Opus 4.x$15.00$75.00
GPT-4.1$2.00$8.00
GPT-4.1 mini$0.40$1.60

A typical Cursor session with full repo context can fire 50–200K input tokens per prompt because Cursor pulls in surrounding files, your edits, and the diff. On Opus, that’s $0.75–$3.00 per single prompt. On Sonnet, $0.15–$0.60. A heavy day of building can easily run $10–$30 in BYO-key spend — multiply across a month and your “just $20 for Cursor Pro” turns into $200–$500.

The honest take: BYO-key is rarely cheaper than Cursor Pro for solo founders. Cursor negotiates volume rates with model vendors and resells access at a healthy discount via the premium-request system. BYO-key makes sense in two cases: (1) you’re a Business customer with strict data-residency requirements that force you to your own Anthropic AWS-Bedrock keys, or (2) you blow through 500 premium requests in week one and need a relief valve until the next billing cycle.

Background agents are metered separately

Cursor’s background agent feature — the “assign a task and let it run” mode that ships diffs autonomously — consumes premium requests at a different rate than interactive editing. Each agent run can fire dozens of model calls in the background as it plans, edits, tests, and revises.

Cursor publishes the agent multipliers separately and counts them against your monthly fast-request budget. In practice, a single non-trivial agent task (“refactor this module to use the new auth helper”) eats 5–15 premium requests of equivalent budget. A founder who fires off three agent tasks a day will burn through the Pro 500-request quota in roughly two weeks.

That’s the unspoken upgrade path: agents push you off Pro into either BYO-key overflow or, if you have a team, the Business tier with multiple seats consolidating quota. If you’re an agent-heavy single-person operation, your effective monthly spend is closer to $40–$80 than $20.

Three realistic monthly bills

Concrete personas to anchor what you’ll actually pay:

Persona 1 — Light user
$20/month flat

A weekend builder using Cursor Pro a few hours per week. Mostly Sonnet, occasional Opus for tricky bugs. Budget consumed: ~150 of 500 premium requests. No agent use. No BYO-key. Bill stays at the sticker price.

Persona 2 — Daily-driver power user
$20/month flat (with discipline)

Solo founder coding 20–30 hours a week in Cursor. Sticks to Sonnet for routine work, drops to Auto / cursor-small for boilerplate, reserves Opus for the 5–10% of prompts that warrant it. Hits ~450 of 500 premium requests by end of month. No agents. Bill stays at $20.

Persona 3 — Agent-heavy founder
$20–$60/month plus possible API overage

Lives in background agents, frequently runs Opus for planning, has Cursor running on multiple repos. Burns the Pro quota in 12–15 days. Either flips on BYO-key for the rest of the month ($30–$80 in Anthropic spend) or upgrades to Business ($40) for the larger pooled budget. Realistic all-in: $50–$80/month.

Most solo founders fall into Persona 2 if they’re paying attention. Persona 3 is the failure mode where you’ve stopped thinking about model selection and the bill drifts upward. The fix isn’t a bigger plan — it’s reading the in-app usage panel and switching defaults back to Sonnet.

Cheaper alternatives if you’re cost-conscious

If $20/month feels steep, or you’ve hit the agent-heavy ceiling and want to spend less, the realistic alternatives are:

  • Windsurf (Codeium): $15/month for the Pro plan. The cheapest credible Cursor alternative, with broadly similar editor UX and autocomplete. Slower model rollout than Cursor — you’ll typically be one Sonnet/GPT version behind. Good fit for cost-conscious solo founders comfortable with marginally older models.
  • Continue.dev: Free open-source extension for VS Code and JetBrains. You bring your own API key and pay the model vendor directly. Total monthly spend with light Sonnet usage can be $5–$15. Much rougher edges than Cursor — no nice agent UI, no “cmd-K-edit” polish — but functional for prompt-and-paste workflows.
  • GitHub Copilot: $10/month for individual. Strong autocomplete, weaker chat and agent features than Cursor. Bundle with a separate Claude / ChatGPT subscription for the chat side and you’re at ~$30 total — not much savings versus Cursor Pro.
  • Sticking with Claude Code or Aider: Free CLI tools where you pay the model API directly. Hard to beat on cost if you’re comfortable on the command line.

For most paying founders, Cursor Pro at $20 still wins on dollars-per-shipped-feature once you account for the time the polished UX saves. We laid out the full tool stack — including where Cursor fits versus Lovable, Claude Code, and the rest — in our best AI tools for solo SaaS founders roundup. The companion Lovable pricing explainer walks through the closest no-code alternative with similar billing dynamics.

Bottom line

Pay $20 for Pro, default to Sonnet, and only escalate to Opus when Sonnet has visibly failed twice in a row. That single discipline keeps 80% of solo founders at the sticker price. If you’re running heavy agent workflows, expect $40–$80/month all-in, and start watching the usage panel weekly. Skip BYO-key unless you have a specific reason — Cursor’s bundled pricing genuinely is cheaper than direct API spend at the volumes most founders work at.

Cursor’s pricing model is honest in one important way: the headline price is real, the overage paths are visible in-app, and there’s no hidden seat math the way some enterprise tools do. The trap isn’t the pricing — it’s your own model-selection habits.

Related reading

Get one SaaS build breakdown every week

The stack, prompts, pricing, and mistakes to avoid — for solo founders building with AI.