Cursor Pricing: The Procurement Model Behind the Flat Per-Seat Number
The Berlin finance partner I sat with last quarter had the Cursor Business proposal in front of her — a clean per-seat number, the headcount, the annual commitment, the Enterprise Agreement-style contract her team knew how to process. Procurement-shape, the easy decision. Three months earlier the same finance team had pushed back hard on a Claude Code line item because the usage-based pricing did not fit their modelling shape. The CTO was about to approve the Cursor proposal in the meeting when the head of engineering walked her through what “500 fast premium requests per user per month” actually meant in the engineering organisation’s workload, which was nowhere near 500 requests per month per engineer for the senior population doing daily Composer Agent work. The Year 1 estimate rose from the clean $9,600 headline to a more honest $32K with the realistic overage modelled in, which the finance team approved without complaint once the math was on paper. The lesson was the same one every Cursor procurement runs into: the per-seat number is honest as far as it goes, and it does not go as far as a finance team modelling from the pricing page assumes.
That is the gap this page is about. Cursor’s pricing is genuinely simpler than Claude Code’s three-billing-mode shape, and the simplicity is a real procurement advantage that the finance function will reward. The simplicity also hides the variable that decides whether the model is right — the fast-premium-request consumption rate at the engineering organisation’s actual workload mix. The pricing-page screenshot will not surface that variable; the engineering organisation has to do it.
This page is the model. Cursor’s four tiers and what each is for, the fast-premium-request mechanics and what triggers overage, the worked-cost example for a 20-engineer team, the Composer Agent cost shape, and where Cursor’s flat subscription wins and loses against Claude Code’s usage-based and Copilot Enterprise’s per-seat models. If you are sizing Cursor for a procurement decision, this is the page that gives you the math the finance partner will sign off on.
The four tiers, briefly
Cursor sells through four tiers, and the procurement decision is which tier covers which population of engineers at what allowance level.
Hobby (Free). A genuine free tier with limited monthly usage — small allowance of fast premium requests, slow-queue access to the premium models, full IDE access. Useful for trial and for solo developers; not viable for professional engineering use at scale. Mention only because the engineering organisation will have some engineers running on Hobby for personal projects.
Pro, $20 per user per month. The individual paid tier. Larger fast-premium-request allowance than Hobby, faster slow-queue access, the Composer Agent included. Pro is the consumer-grade subscription equivalent to Claude Code’s Pro plan. Procurement-wise it has the same friction — it is a personal subscription that procurement cannot easily centralise across an enterprise tenant.
Business, $40 per user per month. The first procurement-grade tier. SSO, central billing, role-based access, the Business model-access tier with the full premium model lineup. Default fast-premium-request allowance lands around 500 per user per month. The Business tier is where most procurement-led Cursor rollouts land in 2026, because it covers the central billing and SSO friction without committing to the Enterprise contract negotiation.
Enterprise, negotiated pricing. Pricing typically lands in the $60-$80 per user per month range depending on the usage allowance, model-access tier, and contracted spend volume. The Enterprise tier adds the audit logging, the centralised admin controls, the SAML SSO with finer-grained policy controls, the higher fast-premium-request allowance, and the standard SOC 2 / ISO 27001 attestations. The contract negotiates the usage allowance and the overage shape, which is the procurement-relevant variable, not the per-seat headline.
The procurement choice between Business and Enterprise turns on two questions: whether the organisation needs the audit and admin surface the Business tier does not provide (regulated industries: yes, most others: not yet), and whether the engineering organisation’s fast-premium-request consumption rate exceeds the Business-tier allowance frequently enough to make the Enterprise upgrade cheaper than the overage. Both questions have to be answered against actual usage data, which procurement does not have before signing — which is the chicken-and-egg problem this page exists to help with.
The fast premium request mechanic, demystified
The variable that decides whether the Cursor cost model is right is the consumption rate of fast premium requests, and the mechanic is worth spelling out because the pricing page glosses it.
A fast premium request is one call to a premium model — GPT-5 family, Claude Opus 4.7, Gemini 2.5 Pro tier — routed through Cursor’s fast inference queue rather than the slower default queue. The fast queue is the experience the engineer expects when working with the tool: low-latency suggestions, snappy chat responses, Composer Agent runs that complete in reasonable time. The slow queue is the fallback when the fast quota is exhausted: same models, much higher latency, materially worse experience that engineers reliably stop using when they hit it.
Each tier ships with a default monthly allowance of fast premium requests — Pro at roughly 500, Business at the same 500 default, Enterprise negotiated higher. Beyond the allowance, the organisation either pays per-request overage at the published rate or the engineer falls back to the slow queue. The published overage rates put one extra fast premium request at roughly $0.04 each, which sounds small until the math runs across an engineering organisation.
The consumption rate is the variable to model. The shape that matters:
- Autocomplete suggestions consume fewer fast premium requests per task because the suggestions are short and use cheaper models by default; the heavy fast-premium consumption is on chat and Composer Agent.
- Cursor Chat consumes one fast premium request per chat turn at the premium model tier. A typical chat session of 5-15 turns burns 5-15 fast premium requests, and a heavy engineer running multiple chat sessions per day consumes more.
- Composer Agent runs consume multiple fast premium requests per agent turn because the agent makes multiple model calls per task. A single Composer Agent run on a meaningful multi-file change can consume 20-50 fast premium requests, depending on the complexity. A senior engineer running Composer Agent heavily can exhaust the Business-tier 500-request monthly allowance inside two weeks of focused use.
The mid-case modelling assumption that has held up in the engagements I have advised: a typical Business-tier engineer doing daily Cursor work consumes roughly 600-1,000 fast premium requests per month, which puts them 20-100% over the default Business allowance. The overage line on the procurement model is real, and the procurement-page-only model that assumes 500/month is accurate does not match the engineering organisation’s actual workload.
Enterprise plan deltas
The Enterprise tier adds five things that matter to the procurement function above the Business tier.
SAML SSO with policy controls. Business-tier SSO is competent; Enterprise adds finer-grained group-based access controls and policy enforcement that the security team will require for regulated workloads. For a 20-engineer team without that requirement, the Business-tier SSO is adequate; for a 200-engineer enterprise rollout in a regulated industry, the Enterprise tier is the floor.
Audit logging. The Enterprise tier produces audit events at the per-user level with SIEM-export capability — authentication, configuration changes, usage. The CISO will require this for regulated workloads; outside regulation, the requirement is softer and depends on the organisation’s internal control posture.
Centralised admin and provisioning. The Business tier gives admins control over the team membership; Enterprise gives admins control over the policy applied to that membership — model access restrictions, data-handling defaults, project-scoping. For a large rollout, the admin surface saves real platform-engineering time over the year.
Higher fast-premium-request allowance. The Enterprise contract negotiates the per-user allowance, typically at 1,000 fast premium requests per user per month as the floor, with higher allowances available against larger contracted spend commitments. The higher allowance flattens the overage volatility the Business tier creates.
Data-handling guarantees. Enterprise provides explicit data-handling guarantees — no training on customer data, configurable retention, the standard SOC 2 / ISO 27001 attestations, BAA availability for the healthcare-customer cases. The Business tier is competent here but the Enterprise tier is the procurement-grade documentation the legal team will want for regulated workloads.
The Enterprise tier is genuinely worth the premium over Business for any organisation with a real compliance requirement, a 100+ engineer rollout, or a heavy-usage population whose overage would exceed the price delta. For a smaller team with no regulatory pressure and lighter usage, Business is the right floor.
Worked-cost example: 20-engineer team
The same shape as the Claude Code worked example so the comparison runs cleanly. Twenty engineers, senior-heavy distribution, Cursor as a primary IDE tool potentially complementing GitHub Copilot or Claude Code.
| Population | Plan | Monthly cost | Annual cost |
|---|---|---|---|
| 8 senior engineers (heavy Composer Agent users) | Business | $40 × 8 = $320 | $3,840 |
| 8 mid-level engineers (moderate Cursor users) | Business | $40 × 8 = $320 | $3,840 |
| 4 mid-level engineers (occasional Cursor users) | Business | $40 × 4 = $160 | $1,920 |
| Base subscription total | $800 | $9,600 | |
| Fast premium request overage (8 senior engineers, ~500 overage/mo each) | At $0.04 per request | ~$160/mo | ~$1,920 |
| Fast premium request overage (8 mid-level engineers, ~200 overage/mo each) | At $0.04 per request | ~$64/mo | ~$770 |
| Spike-month contingency | — | ~$3,000 | |
| Platform engineering integration time | One engineer, 5% allocation | — | ~$7,500 |
| Year 1 total (mid-case) | ~$22,790 |
The mid-case lands at roughly $23K Year 1, which makes Cursor materially cheaper than the equivalent Claude Code mid-case at $52K for the same 20 engineers. The cost gap is real and it is the procurement-relevant signal. The gap closes when the engineering organisation runs heavier Composer Agent workloads (more overage) and widens when the workload is lighter (less overage). The pattern that breaks the model: engineers who exhaust the Business allowance early in the month, fall back to the slow queue for two weeks, and effectively stop using the tool until the next billing cycle. That pattern shows up in usage data as a productivity dip rather than as a billing spike, and the finance model has to account for it as a quality-of-deployment cost rather than a direct cost line.
The high-case version of the same model lands at roughly $45K Year 1, with the difference primarily in the overage line (if all 16 active users exceed their allowance by 50-100% on average) and an Enterprise-tier upgrade ($30K-$50K minimum) for the central admin and higher allowance. The low-case lands at roughly $14K Year 1, assuming light usage and no overage. The honest model spans all three; the budget-meeting ask is the mid-case with the high-case contingency budgeted.
Comparison with Claude Code and Copilot at the licence level
The cost picture at the 20-engineer scale, side by side:
| Tool | Mid-case Year 1 total |
|---|---|
| GitHub Copilot Enterprise | ~$9,400 (flat per-seat) |
| Cursor Business | ~$23,000 (this page’s model) |
| Cursor Enterprise | ~$35,000-$45,000 |
| Claude Code (Pro + Max + API mix) | ~$52,000 (Claude Code pricing page model) |
The shape worth naming. Copilot Enterprise wins on the headline per-seat cost because Microsoft’s procurement engine is built to optimise that number for the broadest possible adoption. Cursor Business sits in the middle — more capable agent than Copilot’s default, less procurement-grade than Cursor Enterprise, materially cheaper than Claude Code at the same population. Claude Code is the most expensive at this scale because the agentic terminal loop is a different shape of product, and the senior population that benefits from it is paying for capability that Copilot and Cursor do not yet match.
The decision is not about the cheapest number. It is about the workload mix the tool actually serves. For a 20-engineer team whose work is predominantly IDE-native and whose senior engineers run Composer Agent rather than terminal-native agents, Cursor Business is the defensible procurement choice. For the same team whose senior engineers do heavy multi-file terminal refactor work, the Claude Code premium is defensible. For a team whose work tilts heavily toward routine business logic and autocomplete, Copilot Enterprise wins on cost and adequacy simultaneously. The Cursor vs Claude Code and Cursor vs Windsurf pieces cover the workflow-fit decision in detail.
Hidden costs the pricing page does not mention
Three lines that finance teams miss on the first Cursor pass.
Composer Agent overage during refactor sprints. Composer Agent runs consume multiple fast premium requests per turn, and a senior engineer running heavy Composer Agent work during a multi-week refactor can exhaust the Business allowance early in the month. The overage line is real, predictable in shape (it tracks the engineering organisation’s refactor calendar), and not captured by the pricing-page allowance. Budget a 25-40% contingency on the senior-engineer line or the model will be wrong on the months that matter.
Model swap costs. Cursor lets engineers choose the model for chat and Composer Agent — GPT-5, Claude Opus 4.7, Gemini 2.5 Pro, the cheaper auto-model. Premium-model selection consumes fast premium requests at the published rate; the auto-model is cheaper but materially less capable on hard tasks. Engineers who default to the premium models for everything consume the allowance faster than engineers who switch deliberately. The cost-discipline conversation worth having: which model defaults are right for which task class, and what the organisation’s standard is. Without that conversation, the model-default behaviour decides the overage rate, and the finance team is reading the consequence rather than setting the policy.
Large context costs. Composer Agent loads large context windows during multi-file refactors, and the context loading consumes fast premium requests at the same per-call rate as the model output. Engineers working on large codebases consume meaningfully more than engineers working on smaller ones, even at the same workflow type. The cost shape correlates with codebase size as well as with workload mix, which the per-seat pricing does not surface and the procurement model has to account for.
When Cursor’s flat subscription stops winning
The vendor-honest version of the same question Cursor’s marketing will not write.
When the workload is heavy on terminal-native agentic loops. Cursor Composer Agent is competent and the agent capability has improved fast through 2025-2026, but the terminal-native agentic loop Claude Code provides is a different shape of product. Engineering organisations whose senior population does heavy multi-file refactor work running through the engineer’s existing shell, makefile, test runner, and deployment scripts will get materially better outcomes from Claude Code at the senior tier despite the higher per-seat cost. The procurement decision is to scope the Claude Code spend to that population rather than try to replace it with cheaper Cursor seats.
When the organisation is at full enterprise scale and Microsoft-shop. At 300+ seats inside a Microsoft Enterprise Agreement, Copilot Enterprise’s volume discounting and the Microsoft compliance integration depth structurally beat Cursor on the like-for-like baseline rollout. The Cursor spend is defensible when scoped to the senior subset or to specific teams whose workflow demands the Cursor experience; it is not defensible as the full-engineering-organisation baseline at this scale and inside this compliance posture.
When the engineering organisation cannot enforce model-default discipline. Cursor’s flat-per-seat economics depend on the engineering organisation establishing model-default policies that keep the fast-premium-request consumption inside the allowance. Engineering organisations that cannot or will not establish those defaults will see the overage line grow until the model breaks. In that case the structural answer is to either move to the Enterprise tier (higher allowance flattens the volatility) or to a different tool whose pricing model handles the discipline gap differently.
How this connects
The AI coding tools hub covers the broader category. The Cursor vs Claude Code piece covers the workflow-fit decision that sits underneath the pricing comparison. The Cursor vs Windsurf piece covers the other IDE-native competitor in the same procurement category. The Claude Code pricing page covers the equivalent cost-modelling exercise for the agentic terminal loop alternative. The AI for engineering teams page covers the team-level shipping gain the Cursor Year 1 spend is being defended against.
The strategy decision upstream of the procurement choice lives at the root hub and the capabilities hub. If your engineering organisation is being asked to buy Cursor without a strategic frame for what the spend is supposed to produce, the procurement model will be defensible at the line-item level and indefensible at the strategic level. Write the strategy frame first; size the Cursor line against it second.
Sources & methodology
- Cursor — Pricing — primary vendor reference for the Free, Pro, Business, and Enterprise tier pricing
- Cursor — Models and pricing documentation — fast premium request mechanics and the per-request overage rate reference
- Anthropic — Claude Code pricing — comparison reference for the Claude Code cost-model alternative
- GitHub Copilot pricing — comparison reference for the Copilot Enterprise per-seat baseline
- Stack Overflow Developer Survey 2025 — adoption-rate data underlying the workload-mix assumptions in the worked example
- Methodology: cost-shape and worked-example figures drawn from fractional CTO procurement engagements (2024-2026) across organisations sizing Cursor and competing AI coding tool spend. Pricing reflects published vendor list pricing as of mid-2026 and will move; treat the worked example as a model template rather than a current-quarter price.
Pricing on Cursor changes quarterly. The fast-premium-request allowances and the per-request overage rates have shifted twice in 2025 alone. Before signing a contract, rerun the worked-cost example against current published pricing and your engineering organisation’s actual consumption data — the model template stays valid; the inputs will need refreshing. If the structure of the model is wrong for your organisation’s shape, send the disagreement and the next refresh will reflect it.