Key Takeaways
- Cursor's documented individual tiers are Pro at $20/mo, Pro Plus at $60/mo, and Ultra at $200/mo, with a free Hobby plan; Teams runs $40/user/mo Standard and $120/user/mo Premium.
- In AIMultiple's 10-task full-stack benchmark, Cursor was the strongest AI code editor at 0.751 combined score but ranked fourth overall behind three CLI agents, and cost $27.90 per task versus $1.03 for the top-scoring CLI.
- Cursor's docs list its own models — Composer 2.5 at $0.50/$2.50 per million input/output tokens, and Grok 4.6 at $2/$6 — priced well below frontier third-party models.
- Forbes reported Cursor's annualized revenue crossed $2 billion and its latest round valued the company at nearly $30 billion, even as an internal January all-hands was titled "War Time."
Cursor is still the best AI code editor you can put in front of a developer who wants to stay in the loop — and it is no longer the cheapest way to get code written. At $20/month for Pro, per Cursor’s own pricing page, it remains the default paid tier for most individual developers. But independent benchmarking now shows command-line agents matching Cursor’s accuracy at a fraction of the per-task cost, and Cursor’s own leadership has publicly reframed the company around building models rather than the editor itself.
That tension is the whole story of Cursor in 2026. Here is what you actually get for the money.
What Cursor is in 2026
Cursor is a fork of VS Code with AI woven through every surface: tab completion, inline edits, and Composer, its multi-file agent. Augment Code’s competitive comparison describes Composer as a mixture-of-experts model trained via reinforcement learning in development environments, and notes it completes most conversational turns in under 30 seconds. That speed is the product — you are paying to stay in flow state while an agent edits alongside you, not to hand off a ticket and walk away.
Cursor’s pricing page lists MCPs, skills, and hooks on every paid individual tier, plus cloud agents and Bugbot — its agentic code reviewer — on usage-based billing. Augment Code also cites Automations, JetBrains support, and an MCP plugin marketplace as recent additions, and notes Composer supports checkpoint rollback and per-file undo but will not automatically revert an entire multi-file change set when a modification fails validation.
Pricing: what the tiers actually cost
Cursor’s marketing page is coy: it shows “$20 / mo.” for individuals and “$40 / user / mo.” for Teams, folding Pro+ and Ultra under one card. The documentation is explicit:
| Plan | Price (per Cursor docs) | Headline limit |
|---|---|---|
| Hobby | Free, no credit card | Limited Agent requests, access to Composer |
| Pro | $20/mo | Extended Agent limits, frontier models, cloud agents |
| Pro Plus | $60/mo | “3x Pro limits on Agent” |
| Ultra | $200/mo | “20x Pro limits on Agent,” priority access to new features |
| Teams Standard | $40/user/mo | SSO, team-wide privacy mode, usage analytics, Bugbot reviews |
| Teams Premium | $120/user/mo | “5x Standard limits on Agent” |
| Enterprise | Custom | Pooled usage, SCIM, audit logs, invoice billing |
Cursor’s docs also list a Start plan at ₹649/mo, tax inclusive, available in India only.
The important caveat: Cursor does not publish how much included usage each dollar buys. Its docs say only that “different models have different API costs, your model selection affects how quickly your included usage is consumed.” When you run out, you can add on-demand usage and “continue at the same API rates with pay-as-you-go billing,” and “requests are never downgraded in quality or speed.” That is a genuinely good overage policy — no throttling, no silent model swap — but it means your real bill is a function of which model you pick, not which plan you bought. Augment Code’s comparison page claims, citing Cursor’s docs, that daily Agent-mode users typically spend $60–$100/month; we could not verify that on Cursor’s own pages, and Augment Code sells a competing product, so treat it as directional. The structural point stands: $20 is the entry fee, not the ceiling.
The model menu is where your budget lives
Cursor’s models documentation splits usage into two pools, and that split is the most consequential thing in its economics. The Cursor Models pool covers in-house options: Composer 2.5 at $0.50 per million input tokens and $2.50 per million output, and Grok 4.6 at $2 and $6, which the docs describe as jointly trained by Cursor and SpaceXAI. “Fast” variants cost roughly double.
The Other Models pool covers Claude, GPT, and Gemini variants at each provider’s API rates, plus a $0.25 per million token Cursor Token Rate on Teams and Enterprise. The Cursor Models pool has more generous included usage — the pricing is engineered to steer you toward Composer, exactly what you would expect from a company whose stated priority is building its own coding model. Route everything through Claude Opus out of habit and your included usage evaporates fast.
The benchmark reality check
AIMultiple ran the most useful head-to-head available: 10 full-stack web development tasks, roughly 600 atomic validation checks per agent, more than 9,600 automated test executions. Cursor was the best AI code editor at a 0.751 combined score on its native Claude Opus 4.6 — and finished fourth overall. Three CLI agents beat it: Opencode at 0.816, Grok at 0.803, and Claude Code at 0.789, all running the weaker Claude Sonnet 4.6.
The cost gap is brutal. AIMultiple measured Cursor at $27.90 per task against $1.03 for Opencode, $1.83 for Claude Code, and $2.03 for Grok, attributing the difference to what editors carry: browser automation, workspace indexing, IDE plugin orchestration, and persistent interaction layers. AIMultiple notes the model differs across categories, so read the cross-category gap as indicative rather than controlled, and subscription pricing masks much of it for real users. Still: a well-orchestrated CLI on a mid-tier model matched a native-Opus editor on full-stack work.
AIMultiple also could not fully automate the editors, because they “frequently require manual approval for certain commands, even when those commands are included in the allowlist.” Whether that is a bug or the entire point depends on how much you trust an agent with your repo — a trade-off we unpack in our comparison of the best AI coding assistants.
Where Cursor still wins
Two things. UI-heavy work: AIMultiple noted editors post near-perfect UI scores and carry more built-in debugging tooling, and observed Cursor opening a browser window during tasks. And human-in-the-loop control — inline diffs, visible intermediate steps, the ability to intervene mid-execution. AIMultiple frames the split cleanly: CLI tools optimize for system-level automation, editors optimize for human-in-the-loop productivity.
That second advantage has an expiry date. Business Insider reported that new data from Cursor shows the share of AI-generated code changes reaching production without a separate manual review step has jumped over the past six months, and that Cursor says AI-generated code is surviving at higher rates than before. If nobody reads the diff, the inline-diff advantage stops mattering — precisely the strategic hole Cursor is climbing out of. Our head-to-head on Cursor, Claude Code, and Copilot breaks down which workflows survive that shift.
The company behind the editor
This matters before you commit a team to a seat count. Forbes reported that on January 5, Cursor employees returned to an all-hands with a deck titled “War Time,” after staff testing Anthropic’s Opus 4.5 realized developers no longer needed to review every line of output. The new mandate, per Forbes: “P0 #1” — “Build the best coding model.” Not the best wrapper.
The business is not collapsing. Forbes reported Cursor began 2025 at roughly $100 million in annualized revenue, surpassed $1 billion by November, and — per a source familiar with its finances — crossed $2 billion, doubling in three months, with its latest round valuing the company at nearly $30 billion. But Forbes also reported Claude Code crossed $1 billion in annualized revenue within six months and hit $2.5 billion, surpassing Cursor, and that Ramp data shows Cursor’s adoption rate among businesses buying AI products declining slightly. For the broader model-layer picture, see our ChatGPT vs Claude vs Gemini breakdown.
Who should buy what
Free Hobby is a real trial, not a demo — limited Agent requests plus Composer access is enough to judge the editor in an afternoon. Pro at $20 fits developers who use the agent in bursts and still do most of their own typing. Pro Plus at $60 is the honest tier for daily agent users. Ultra at $200 only makes sense once you have measured that you exceed Pro+ limits. For teams, the jump from $40 to $120 per seat buys 5x Agent limits — decide by measured usage, not by job title. Worth pricing against it: Claude Code for CLI throughput, GitHub Copilot if your org lives in GitHub, and Windsurf for a comparable editor experience.
What happens next
Watch three things. Whether Composer’s next version closes the accuracy gap with frontier models — Cursor’s whole margin story depends on you choosing its in-house pool. Whether Cursor publishes concrete included-usage figures, since that opacity is the most common pricing complaint. And enterprise mix: Forbes reported Cursor is prioritizing large-enterprise contracts precisely because they are more stable than consumer subscriptions, a sensible hedge if individual developers keep drifting to CLI agents.
Quick poll
What would make you leave Cursor?
AIMultiple measured Cursor at $27.90 per benchmark task versus $1.03 for the top-scoring CLI agent, Opencode.
FAQ
How much does Cursor cost in 2026? Per Cursor’s docs: Pro $20/mo, Pro Plus $60/mo, Ultra $200/mo, plus a free Hobby plan. Teams seats are $40/user/mo Standard and $120/user/mo Premium; Enterprise is custom-priced.
Will I pay more than $20 a month on Pro? Possibly. Cursor’s docs say that once you exceed included monthly usage you can add on-demand usage at standard API rates, and that pricier models burn included usage faster. Cursor does not publish a dollar figure for included usage.
Is Cursor more accurate than Claude Code? Not in AIMultiple’s benchmark. Cursor scored 0.751 on native Claude Opus 4.6 while Claude Code scored 0.789 on the weaker Sonnet 4.6, with two other CLI agents higher still.
Is Cursor’s own model worth using? Cursor’s docs price Composer 2.5 at $0.50/$2.50 per million input/output tokens — far below frontier third-party rates — and the Cursor Models pool has more generous included usage. Budget-conscious users should test it first.
