"AI programming" sounds like one more subscription—but real monthly bills hide in three places: stacked Pro plans, API overage after quotas, and execution environments for Agents / iOS builds. Using August 2026 list prices, this guide gives a reproducible formula and four developer profiles so you can put a number in your budget.
Monthly cost formula: three layers
Writing AI programming spend as one line beats guessing which Pro to buy:
Monthly total (individual / small team)
Monthly ≈ Σ subscriptions + API inference (incl. overage) + execution env (Mac / CI) + team seats × headcount
Most people count only the first term; Agent users see the second spike; Apple platform devs need the third.
2026 subscription price table
Typical individual/standard monthly list prices in August 2026 (USD, ex tax; enterprise pricing separate). Mobile IAP may cost more.
| Tool | Tier | ~Monthly | Typical use |
|---|---|---|---|
| Cursor | Pro | $20 | AI IDE, Tab completion, limited Agent |
| Cursor | Business | $40/user | Team policy, central billing |
| GitHub Copilot | Individual | $10 | IDE completion (many editors) |
| GitHub Copilot | Business | $19/user | Org policy and audit |
| Claude (web + Code) | Pro | $20 | Claude Code base usage pool |
| Claude | Max 5x | $100 | Heavy terminal Agent |
| Claude | Max 20x | $200 | All-day parallel Agents |
| ChatGPT | Plus | $20 | General chat, some Codex |
| ChatGPT | Pro | $200 | High-quota Pro models |
| Windsurf | Pro | ~$15 | AI IDE competitor |
Key point: subscriptions cover in-product AI quotas—they are not unlimited API. Cursor fast/slow requests and Claude's 5-hour rolling windows push you to pay-as-you-go or upgrade tiers. For Max tier choice see Is Claude Code Max worth it.
Layer two: API and overage
Custom Agents, automation pipelines, or "extra usage" after quotas switch billing to tokens—the same core formula as our OpenRouter API pricing guide:
Per call ≈ (input tokens ÷ 1M × input rate) + (output tokens ÷ 1M × output rate)
2026 flagship reference rates (OpenRouter / vendor list, USD per million tokens):
- Claude Sonnet 4.6: ~$3 / $15
- GPT-5.4: ~$2.5 / $12
- Gemini 3.1 Pro: ~$2 / $12
In Agent workloads input tokens often dominate (repo reads, history, tool results). Twenty daily Sonnet sessions at 50K input + 2K output ≈ $15/day ≈ $450/month inference alone—why many buy Max first until subscriptions cap out.
Subscription vs API rough split
- Monthly inference < $100 + heavy Claude Code terminal → Pro / Max subscription
- Monthly inference > $200 + multi-model routing → pure API (OpenRouter)
- Middle ground → subscription + some overage is typical
Four profiles: real monthly bills
Figures below are tool-layer estimates (ex tax); execution environment extra.
Profile A: hobby / side project (light)
- Tools: Copilot Individual $10 or Cursor Pro $20
- API: ~$0
- Total ~$10–30/month
Profile B: full-time indie (medium)
- Tools: Cursor Pro $20 + Claude Pro $20
- API overage / small Agent: $30–80
- Total ~$70–120/month
Profile C: heavy Agent solo
- Tools: Claude Max 5x $100 + Cursor Pro $20 (or Copilot $10)
- API overage: $20–50
- Total ~$120–170/month; Max 20x + ChatGPT Pro can hit $400+/month
Profile D: 5-person team
- Copilot Business 5×$19 = $95, or Cursor Business 5×$40 = $200
- Shared API / Agent backend: $200–800
- Optional cloud Mac CI (1 node): $50–150
- Total ~$350–1,150/month; outward SaaS often higher
| Profile | Subscriptions | API | Monthly (tools) |
|---|---|---|---|
| A light | $10–20 | $0 | $10–30 |
| B medium | $40 | $30–80 | $70–120 |
| C heavy solo | $120–420 | $20–50 | $120–470 |
| D team 5 | $95–200 | $200–800 | $350–1,150+ |
Hidden costs people miss
- Duplicate subscriptions: Copilot + Cursor + Claude Pro overlap—you pay for three but use one workflow.
- Free trial traps: 14-day Pro auto-renews; team seats multiply by headcount.
- Top-up fees: OpenRouter card top-up ~+5.5%—budget it even on small loads.
- Execution environment: Agents still need Mac
xcodebuild, signing, TestFlight—see cloud Mac / iOS toolchain costs, often opened the same day as LLM tools. - Context bloat: unread compressed history and whole-monorepo reads cost more than switching flagship models.
Self-audit: where you land
Ten minutes on this checklist beats gut feel:
- List all active subscriptions (annual ÷ 12)
- Export last month's API
usageor platform bills - Sum "extra usage" and BYOK overage lines
- Add Mac VPS / cloud Mac if you run iOS/macOS pipelines
- Formula: Σ subs + Σ API + Σ env, then ÷ team size for per-person
Quick takeaways
- Subscription-only math undercounts 30–70% for Agent users
- Light solo $10–30; full-time solo often $70–170
- Teams: seats × headcount + shared API, not subscription price alone
- Cut costs: drop duplicates → compress context → model routing before downgrading models
FAQ
How much do student / OSS discounts save?
Copilot and some IDEs offer free or discounted tiers for students or maintainers—Profile A can drop to ~$0–10/month, but enterprise compliance usually can't substitute personal education accounts for Business seats.
What finance teams want for reimbursement?
Beyond invoices: last month's usage screenshot plus feature mapping helps avoid "duplicate Cursor and Claude" rejections. API split by project beats one lump sum.
Web-only frontend—need Claude Code Max?
Short tasks, small repos: Pro + Cursor usually enough. Long daily Agents, multi-file refactors, parallel subtasks → evaluate Max 5x.
AI writes code; cloud Mac runs builds
Model subscriptions answer "who writes code"; Apple signing and TestFlight still need macOS. Zilmac cloud Mac pairs with Cursor / Claude Code workflows and moves xcodebuild off your laptop.
Run the full Apple toolchain without a physical Mac. — View cloud Mac plans