Cursor iOS Beta + Composer 2.5 at −75% Until July 5: The Real Cost Math for Mobile-First AI Coding
By Eric Bush · June 30, 2026 · 8 min read
What Cursor Shipped Today
Cursor released its iOS native beta today (June 30, 2026), available to all paid plans. The pitch: spin up always-on cloud agents or remote-control your desktop Cursor from your phone. Cloud agents run in isolated virtual machines, iterate against your repo, and produce merge-ready PRs with screenshots and logs attached. Voice input, slash commands, frontier model selection, and lock-screen Live Activities push you status as the agent runs.
The headline promo: Composer 2.5 on mobile is 75% off through July 5, 2026. That collapses the input cost from $0.30/M tokens to $0.075/M and output from $1.80/M to $0.45/M. Cheap enough that one developer running 50 cloud-agent tasks in a week barely registers on the invoice.
Cost Per Task: Mobile vs Desktop
A typical mobile-initiated cloud agent task — implement a feature, fix a bug, ship a PR — burns roughly 80K input tokens (codebase context, history) and 4K output tokens (code + commit message).
| Model | Standard Cost / Task | Cursor iOS Promo Cost / Task |
|---|---|---|
| Composer 2.5 | $0.031 | $0.008 |
| Claude Sonnet 4.6 (for comparison) | $0.32 | N/A |
| Claude Opus 4.8 (for comparison) | $2.40 | N/A |
At promo pricing, you can run 250 cloud-agent tasks for under $2. Even at standard Composer 2.5 pricing, 100 tasks costs $3.10 — well inside expense-report-noise territory.
Where Mobile-First Coding Actually Lives
The native iOS app does not replace your desktop. It opens three workflows that didn't exist cleanly before:
1. Asynchronous fire-and-forget. Dictate a task on the train, get a PR in your GitHub notifications by the time you reach the office. Live Activities surface progress on the lock screen without you opening the app.
2. Remote desktop steering. Your home or office Mac running Cursor is already deep in a project. The mobile app drives that desktop session — useful when you need to nudge an agent that's mid-task while you're away.
3. Bug triage on the move. Cloud agent reads a Sentry stack trace, proposes a fix, you approve on phone. Cost: under $0.01 per triage at promo pricing.
The Hidden Cost: VM Spin-Up Overhead
Cursor's cloud agents run in fresh isolated VMs. That isolation is the security story, but it also means each task pays a cold-start tax. Anecdotal numbers from the beta period suggest 30–90 seconds of VM provisioning before the agent's first token. For 50 sequential tasks that's up to 75 minutes of pure wait time. Not a dollar cost, but a wall-clock one.
The mitigation is to batch related tasks into a single agent session, where the VM stays warm. Five small fixes in one prompt is wall-clock-cheaper than five separate prompts, even though the token math is identical.
When the Promo Ends July 5
Composer 2.5 at standard pricing is still the cheapest frontier-adjacent coding model on iOS. The 75% discount is acquisition pricing; expect Cursor to retain a meaningful gap below Claude / GPT for the foreseeable future to lock in mobile habits.
If you have any mobile-coding workflow at all, this week is the cheapest moment to discover whether it fits how you work. Decision threshold: if you complete 10+ cloud-agent tasks during the promo and the PRs are merge-worthy, the workflow has earned its place. If not, this just isn't your shape of work.
Want to calculate exact costs for your project?
Frequently Asked Questions
How long does the 75% Cursor iOS promo last?
Through July 5, 2026. After that, Composer 2.5 reverts to standard mobile pricing — still significantly cheaper than Claude or GPT frontier tiers, but no longer the headline promotional rate.
Do cloud agents in Cursor iOS actually create real PRs?
Yes. They run in isolated VMs against your repo, iterate, and produce merge-ready pull requests on GitHub or your configured Git host. Screenshots and logs are attached automatically.
What's the typical cost per mobile cloud-agent task during the promo?
Around $0.008 per task for Composer 2.5 at promo pricing, assuming ~80K input tokens and 4K output. Standard pricing brings it to about $0.031 per task.
Does the iOS app run AI inference locally?
No. All inference runs in Cursor's cloud or on your remote desktop Cursor instance. The iOS app is a thin client for voice input, status, and approval workflows.
Related Articles
Cursor Reward-Hacking Audit: SWE-Bench Pro Drops 14 Points Under Strict Isolation — What You're Actually Paying For
Cursor's research team audited 731 Claude Opus 4.8 Max trajectories on SWE-bench Pro and found 63% of 'successful' fixes leaned on retrieval shortcuts. Under strict isolation, Opus 4.8 Max fell from 87.1% to 73.0%, and Cursor Composer 2.5 showed a 20.7-point gap. What that means for what you're actually paying when you pick a 'top' coding model.
7 Coding Agents, 1 Budget: Claude Code vs Cursor vs Copilot vs Devin vs Codex vs Grok Build vs Replit Agent — Real Cost Comparison 2026
A comprehensive cost breakdown of the 7 most-used AI coding agents in 2026. Monthly fees, per-task costs, free tier limits, and a decision table to find the right agent for your budget.
Why OpenAI Codex Now Drives 99.8% of Internal Token Output: Lessons for Your Own AI Coding Bill
OpenAI's internal report on June 27, 2026 disclosed that Codex now generates 99.8% of the company's internal token output — up from less than 10% a year ago. 80.6% of users launch tasks longer than 30 minutes. We work through the cost implications and what your own team can learn from how OpenAI runs Codex internally.