跪拜 Guibai
← Back to the summary

AI Coding Subscriptions Can Cost Two MacBook Pros a Year

I Did the Math: AI Coding Tools Burn Two MacBook Pros a Year

Author: kyriewen

On October 13, Cursor quietly changed its plans.

New users landing on the pricing page no longer see the familiar $20 Pro; instead they see four tiers: $8, $30, $100, and $200. Existing subscribers renew at the old price, while new users are funneled into a new pricing system: the mid-tier went from $20 to $30, a 50% increase, and it didn't even make trending. I went through five other tools as well, and after calculating the annual total, here's the spoiler: a heavy user running Cursor Ultra and Claude Max 20x simultaneously costs $4,800 a year, which at a 7.2 exchange rate is over 34,000 RMB, exactly two MacBook Pro 14s. Whether that money is well spent, you decide after seeing the price table.

The $20 Cursor Is Gone

Let's talk about the change. Since October 13, the new Cursor user plans are:

Old Pro users renew at $20, but new users need to pay $10 more to get back to that tier. The official line is "more flexible tiers," but effectively it's a 50% price increase on the middle tier.

And this move was quiet. No announcement blitz, no trending topics, and the sparse Zhihu discussions drowned in the traffic of "AI tool comparison reviews." Compared with the collective tightening in June that "ended the monthly subscription era," this time only Cursor moved, and it was much more hidden.

Claude Goes the Other Way

On October 7, Anthropic did the opposite: Max subscriptions started giving away API credits.

After subscribing for 7 days, you link your organization in the Console to claim them. The credits can only be used on API and CLI tools like Claude Code. In plain language: spend $200 on Max 20x and it gives you $200 back in API credits, which halves the subscription cost, provided you're working through Claude Code.

One quietly raises prices, the other openly gives away. Copilot stays honest; the $10 Pro price hasn't moved.

OpenAI Also Moved, and It's Sneakier

Codex doesn't have its own subscription; it's bundled into ChatGPT: Go $8, Plus $20, Pro $100/$200/$500. The entry is cheap, and costs go up from there.

The thing to watch is tomorrow: starting October 30, qualifying existing Pro subscriptions will be switched to new, lower quotas. The price hasn't dropped a cent, but the quota is cut first. The official wording calls it "adjusting quota structure," which translates to: a price increase without saying so, and a cut without negotiation. New subscriptions start with the new quotas from day one, without even a transition period.

Cursor's 50% increase at least has a clear label; OpenAI's move is "you renew and find the same money buys less." Which one is more infuriating? Vote in the comments.

Six Tools, One Table

Tool Free tier Entry Mid High
Cursor $0 Plus $8 Premium $30 Super $100 / Ultra $200
Claude $0 (limited Sonnet) — Pro $20 Max 5x $100 / Max 20x $200
Codex (ChatGPT subscription) Free limited Go $8 Plus $20 Pro $100 / $200 / $500
GitHub Copilot $0 Pro $10 Pro+ $39 Business $19/user
Tongyi Lingma Free basic Personal Pro ¥59/month Enterprise Standard ¥79/user Enterprise Exclusive ¥159/user
Doubao MarsCode Free — — —

Four international and two domestic, prices from $0 to $500 forming a complete hierarchy of contempt.

The Annual Bill

Unit prices show nothing; do the math annually. Three typical usage patterns:

Combination Monthly cost Yearly cost Equivalent
Heavy dual-open: Cursor Ultra + Claude Max 20x $400 ¥34,560 Two MacBook Pro 14s
Heavy single-open: Codex Pro $200 ¥17,280 One MacBook Pro 14
Mainstream dual-open: Cursor Premium + Claude Pro $50 ¥4,320 One iPad 10
Mainstream mix: Cursor Premium + Copilot Pro $40 ¥3,456 One iPhone SE
Light: Cursor Plus / Codex Go $8 ¥691 Two Haidilao hotpot dinners
Zero-cost: MarsCode + Tongyi Lingma basic ¥0 ¥0 Priceless

You can see: high cost is never one vendor's fault; it's stacking's fault. By itself, Cursor at $30 a month is okay, Claude at $20 isn't expensive, Codex Plus at $20 certainly isn't, but the agentic-era workflow is about running multiple tools: Claude Code writes core logic, Cursor reviews diffs and edits the UI, Copilot provides fallback autocomplete in the IDE. Stack three subscriptions and you're out nearly a hundred dollars a month.

A 50% price hike forces you to pay an extra $10; stacking forces you to pay an extra $200. The latter is the real bill assassin.

Reimbursed vs. Self-Paid: Never the Same

Before the math, ask: who pays?

In big companies, AI tool costs go through procurement; Business/Enterprise tiers are allocated per head, and individuals pay nothing. The $39 Copilot Enterprise and the free tier feel almost identical. Reimbursed users never look at price tables.

Self-paid users are the price-sensitive group: independent developers, freelancers, students, and the laid-off doing side gigs. For them, $30 vs. $200 is not a "tier difference"; it's "whether I can afford milk tea this month."

So any post about "X tool costs 40 grand a year" always splits the comments into two camps: one says "company pays, don't care," the other says "self-paid here, can't afford it." They're not arguing about price; they're arguing about position: the reimbursed can't understand every cut the self-paid feels, and the self-paid wants to block anyone who says "it's not that expensive."

This conclusion is split for the two groups:

Some Will Say

"The free tier isn't usable, is it?"

It depends on how you use it. If you're a heavy user, one prompt in Claude Code that generates an entire component, then you review and iterate, a free tier won't last an afternoon. Claude's free tier only has Sonnet with limits; Cursor's free version is mainly autocomplete, and agent mode prompts an upgrade after two rounds.

But the zero-cost crowd has more room than you'd think: Doubao MarsCode is still free (as of the day I collected data, no paid announcement), and Tongyi Lingma's basic tier is free. Tab completion, single-file generation, and Q&A are enough for light work. Domestic developers actually have a complete free stack to live on.

So the answer splits in two: light usage, the free tier is genuinely enough; heavy usage, the free tier is an illusion.

"Is annual payment worth it?"

Claude Pro yearly is $200, about $17/month, saving 15% vs. monthly. Cursor's Plus tier also offers a yearly discount. If you're sure you won't switch tools for a year, annual billing saves real money. But start with monthly for two months; tool generations change faster than a girlfriend switches lipstick shades, and locking in for a year carries risk.

"Is the Max API credit a trick?"

No, this one is real. A $100 subscription includes $100 in credits, which is Anthropic's admission that heavy users' money goes to API usage as the bigger line item. As long as you work through Claude Code, this is free money. There's a condition: after 7 days of subscription, go to the Console, link your organization, and don't forget.

My Verdict

Here's an action list you can copy:

  1. Light usage (autocomplete + single-file generation): MarsCode or Tongyi Lingma basic, ¥0
  2. Medium usage (a few agent rounds per day): pick one of Cursor Premium $30, Claude Pro $20, Codex Plus $20
  3. Heavy usage (Claude Code as main): go straight to Max 5x $100 and claim the API credit, which effectively makes the subscription $0
  4. Don't dual-open more than three; the third subscription is 90% wasted money

Finally, a poll: how much do your AI tools cost per month? Report your number in the comments; I'll calculate a median in the next post and see just how much we frontend devs can burn.