Cursor's pricing changes generated one of the biggest backlashes the AI developer tools space has seen. In late 2025, the company quietly switched from a request-based model to credit-based billing - catching most users completely off guard. Teams that had budgeted around predictable $20/month subscriptions suddenly saw $500+ in overage charges within days.
The community reaction was swift and loud. Developers flooded forums, Reddit, and Trustpilot with complaints. Cursor eventually apologized, rolled back some charges, and issued refunds for affected users. But the underlying problem never went away: most teams still have no clear visibility into what Cursor actually costs them.
If you use Cursor on a team, this post breaks down how the pricing actually works now, where the visibility gaps are, and what you can do about it.
How Cursor pricing actually works
Cursor offers three subscription tiers:
| Plan | Monthly price | What's included |
|---|---|---|
| Free | $0 | 50 slow premium requests per month |
| Pro | $20/user | 500 credits per month, unlimited completions |
| Business | $40/user | 500 credits per month, admin controls, enforced privacy |
The key shift is from requests to credits. Under the old model, Pro users got 500 "fast requests" per month - one request, one count, regardless of model. Under the new model, different models consume different amounts of credits per request.
This matters because not all models cost the same. A request using Claude Sonnet 4 costs roughly 2.2x more credits than the cheapest available option. If your team defaults to premium models (which Cursor often selects automatically), those 500 credits drain much faster than 500 requests used to.
In practice, a Pro subscription now gets you roughly 225 Claude Sonnet requests instead of the old 500 fast requests. That is less than half the effective usage for the same $20/month.
When your credits run out, Cursor switches to on-demand billing. Each additional request is charged at a per-model rate, billed directly to your payment method. There is no hard stop - usage continues and charges accumulate until the end of your billing cycle.
The visibility problem
Here is where things get painful for teams. Cursor does not provide a clear, granular cost breakdown in its UI. You cannot easily answer questions like:
- Which developer on my team used the most credits this month?
- How much did we spend on Claude Sonnet vs. GPT-4o this week?
- What is our daily spend trend - are we on track to blow our budget?
On-demand usage charges accumulate silently. There is no real-time cost indicator showing you that you have moved past your included credits and are now paying per-request. One user reported discovering $60 in overage charges across 2.5 weeks with no warning or notification from Cursor.
The problem compounds with model selection. Developers reported that "expensive thinking model calls were sprinkled across many tiny tasks" - quick questions and small edits that could have used cheaper models but defaulted to premium ones. This kind of waste is completely invisible without external analysis. You do not realize your team is burning through credits on autocomplete suggestions routed to a $15/million-token model until the bill arrives.
What teams are spending
The numbers from the community paint a clear picture of how quickly costs escalate under the new model.
Individual developers regularly report $10-20/day in overages once their credits run out. For a developer who codes 20 working days per month, that is $200-400 on top of the $20 subscription - turning a $20/month tool into a $220-420/month tool.
Team-level costs are worse. One widely-cited case involved a team whose $7,000 annual subscription was effectively depleted in a single day due to heavy usage of premium models. Teams with 10+ developers report their monthly Cursor bill rising from a predictable $500/month (25 seats at $20) to $1,200-3,000/month once overages are factored in.
The dissatisfaction is measurable. As of early 2026, 64% of negative Trustpilot reviews for Cursor cite billing issues - not bugs, not performance, not features. Billing.
| Scenario | Old model (requests) | New model (credits) |
|---|---|---|
| Light user, 200 requests/month | $20/mo (within plan) | $20/mo (within plan) |
| Moderate user, 400 requests/month | $20/mo (within plan) | $20/mo + ~$30 overages |
| Heavy user, 800 requests/month | $20/mo + some overages | $20/mo + ~$150 overages |
| 10-person team, mixed usage | ~$200/mo predictable | $200/mo + $400-1,200 overages |
The pattern is consistent: light users are fine, but moderate-to-heavy users are paying significantly more than they were under the old pricing.
How to track your real Cursor costs
If your team is on a Cursor Enterprise plan, you can connect to Grafient via the Cursor Admin API key and get the visibility that Cursor's own UI does not provide.
Here is what that gives you:
Per-model cost breakdowns. See exactly how much your team is spending on Claude Sonnet, GPT-4o, and every other model Cursor routes to. This is the single most important piece of data for understanding where your budget goes - and it is the one thing Cursor does not surface clearly.
Daily spending trends. Track your team's Cursor spend day by day, not just as a monthly total. A daily view lets you spot when a new hire starts burning through credits, when a sprint pushes usage higher than expected, or when a model routing change silently increases costs.
Budget alerts. Set a daily or monthly spend threshold and get notified before overages hit. Instead of discovering a $300 overage at the end of the month, you get an alert on day 8 when spending starts trending above your target.
To set this up:
- Go to your Cursor Enterprise dashboard and create an Admin API key under Settings > Advanced > Admin API Keys.
- In Grafient, navigate to Integrations, click Add Integration, and select Cursor.
- Paste your Admin API key and save.
Grafient will backfill your historical data and sync daily going forward. The API key is read-only and encrypted - Grafient never performs any actions on your Cursor account.
Note: The Cursor Admin API is currently only available on Enterprise plans. If you are on Pro or Business, the Admin API is not accessible, though this may change as Cursor expands its API offering.
Tips for managing Cursor costs
Even without external tooling, there are steps you can take to keep Cursor costs in check.
Understand your plan limits before the billing cycle. Know how many credits your team gets, and roughly how many requests that translates to for the models you use. If your team primarily uses Claude Sonnet, 500 credits means roughly 225 requests - plan accordingly.
Set team-level budget awareness. Make sure every developer on your team knows that credits are not unlimited, and that on-demand charges kick in silently. A five-minute team standup about Cursor billing can prevent hundreds in unexpected charges.
Compare Cursor AI costs against direct API costs. Cursor is a convenience layer on top of model providers. The per-request cost through Cursor is typically higher than calling the API directly. For some workflows - particularly batch or background tasks - it may be cheaper to use the API directly and reserve Cursor for interactive coding.
Be intentional about model selection. If Cursor lets you choose a model, default to the cheapest option that gets the job done. Reserve premium models for tasks that genuinely need them - complex refactors, multi-file reasoning, or difficult debugging. Autocomplete and simple edits rarely need the most expensive model.
Get visibility across all your AI spend
Cursor is just one part of most teams' AI spend. If you are also using Anthropic's API directly, OpenAI, Google AI, or other providers, the same visibility problem applies everywhere - fragmented billing data, no unified view, no cross-provider alerting.
Grafient tracks Cursor costs alongside all of your other AI providers in a single dashboard. One view for your entire AI budget, with daily breakdowns, model-level detail, and alerts that catch cost spikes before they become billing surprises.
Start tracking your AI costs - connect your first integration in under two minutes.