The Free plan, the $20 Pro tier, and Business at $40/seat — plus the part most founders trip on: what a “premium request” actually is, and when your BYO-key API bill quietly takes over. Numbers verified against cursor.com/pricing.
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.
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.
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:
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.
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:
| Model | Input / 1M tokens | Output / 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.
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.
Concrete personas to anchor what you’ll actually pay:
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.
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.
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.
If $20/month feels steep, or you’ve hit the agent-heavy ceiling and want to spend less, the realistic alternatives are:
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.
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.
The stack, prompts, pricing, and mistakes to avoid — for solo founders building with AI.