Scale · founder · 9 min read

The Vibe Coding Pricing Calendar: Every Dated Change We Can Verify

Every dated change to what AI coding tools cost you, with the vendor page we read it on. Now tracking Claude Code and Codex too. Updated Sep 18.

Updated September 18, 2026 ·

pricing budget cursor github copilot gemini claude code codex models

Most pricing articles tell you what things cost today. That number is wrong within about six weeks, and the article never says so.

This page does something different. It lists price changes that are already scheduled, with the date each one lands and the vendor page we read it on. A dated future event stays true until it happens. A snapshot of current prices starts rotting the moment it publishes.

Seven changes are on the calendar right now. Three have landed.

The calendar

Date	Tool	What changes	Sticker price after
Sep 1, 2026	GitHub Copilot	Landed. Promotional included credits expired. Business 3,000 → 1,900. Enterprise 7,000 → 3,900.	Unchanged: $19 / $39
Sep 7, 2026	Cursor	Legacy Enterprise Auto flat rate ends. Auto bills at each routed model’s list price.	Unchanged: negotiated
Sep 14, 2026	Claude Code	Landed. Temporary +50% weekly limit boost expired; permanent +25% began. Net 17% fewer weekly units than Sep 13.	Unchanged: $20 Pro
Sep 28, 2026	GitHub Copilot	Code review default effort goes Lite → Balanced. More effort, more tokens, more credits drawn.	Unchanged: $19 / $39
Oct 1, 2026	GitHub Copilot	Existing card/PayPal Business and Enterprise customers charged upfront per seat each cycle.	Unchanged: $19 / $39
Oct 14, 2026	OpenAI Codex	GPT-5.5 removed from Codex, ChatGPT and ChatGPT Work on all plans. API keeps it. Migrate to 5.6 Sol or Astra.	Unchanged: $20
Dec 31, 2026	Gemini 3.7 Flash	Introductory API rate ends. Input and output roughly double.	Unchanged: n/a (API)
Sep 1, 2026	Claude Sonnet 5	Cancelled Aug 10. The $2/$10 introductory rate is now permanent.	Unchanged: $2 / $10

Notice the last column. In every single row, the advertised price is identical before and after. That is not a coincidence, and it is the whole reason this page exists.

September 14 and October 14: the bundle shrinks, twice

Two entries added to this calendar on September 18, one from each of the two vendors most likely to be in your stack.

Claude Code lost 17% of its weekly limits on September 14. A temporary 50% boost that had run since May expired on the 13th; a permanent 25% increase over the original baseline started on the 14th. Both percentages are real, measured against different baselines. Against what you actually had on September 13, it is a cut. Anthropic’s first announcement led with the 25%, developers objected, and Anthropic deleted and reposted with “a 17% reduction in weekly limits” stated outright — a same-day correction, which is more than most rows on this table got.

Codex loses GPT-5.5 on October 14. OpenAI is retiring it from ChatGPT, ChatGPT Work and Codex across every plan. The API is explicitly exempt. If you have gpt-5.5 selected in a workspace default, a custom agent, a scheduled task or a hardcoded script, it stops resolving on the 14th — and scheduled tasks fail quietly.

These two belong on the same calendar for the same reason as everything else here: the invoice does not move. Full breakdown in two AI coding plans shrank this month.

September 28: the default gets more expensive

This one is new to the calendar as of today, and it is the subtlest entry on it.

GitHub is changing the default effort level for Copilot code review from Lite to Balanced. That’s it. That’s the whole announcement, filed under product policy rather than pricing.

Effort level determines how much work the model does per review. More work is more tokens, and tokens are what draw down the credit pool that just got 37% smaller for Business seats. So on September 28, a team that changes nothing will consume more of a smaller allowance, at an unchanged seat price, because a dropdown moved.

GitHub does offer an out, and it requires action: if you explicitly select Lite in your repository or organisation settings before September 28, GitHub says it will respect that and won’t switch you. Leaving the setting on Default is what opts you in.

We have argued on this site for months that the price is no longer where the money moves. This is the clearest single example we have found: a cost increase with no number attached to it anywhere.

October 1: pay before you use

Also new to the calendar. GitHub is reopening Copilot Business and Enterprise signups to credit card and PayPal customers, with stronger account vetting, and changing when money changes hands.

New seat assignments now require payment for each seat before the user gets access. From October 1, that upfront charge extends to existing Business and Enterprise customers paying by card or PayPal, applied at the start of each billing cycle. Included usage may be prorated across the month to match seat proration.

GitHub states directly that Business and Enterprise prices are not changing, and that spend controls, usage tracking and additional credit purchases all remain. What changes is your working capital: you now fund the seats before you consume anything, and revoking a seat still produces no prorated refund.

For a 50-seat Business org that is roughly $950 moving a cycle earlier. Not a crisis. Worth knowing before it appears.

September 1: the Copilot cushion expired

When GitHub moved Copilot to usage-based billing on June 1, existing Business and Enterprise customers got three months of elevated included credits. GitHub’s billing docs define that window as “June 1 – September 1, 2026.” It closed on schedule, with no extension and no announcement on the day.

Business seats drop from 3,000 included credits to 1,900. Enterprise drops from 7,000 to 3,900. Credits are worth exactly $0.01 each, so that’s $11 and $31 per seat of included usage disappearing — a 37% and 44% cut. Seat prices don’t move.

Two details make this worse than the numbers suggest. Credits are pooled at the billing-entity level, so no individual developer gets a warning that the team is running dry. And additional usage is on by default — when the pool empties, consumption doesn’t stop, it bills to the org at published rates with no automatic fallback to a cheaper model. A 20-seat Business org with flat usage goes from $380 to roughly $600 a month. We covered the arithmetic in your Copilot bill goes up on Tuesday.

What to do before your next billing cycle closes: set a budget in your org’s billing settings, and decide deliberately whether additional usage should stay enabled. The default answer is yes, and nobody chose it.

September 7: Cursor’s last flat rate

Cursor moved individual and Teams plans to routed-model pricing on August 24. Enterprise customers on legacy Auto arrangements got a two-week extension. Cursor’s own docs are unambiguous: “Until September 7, 2026, Enterprise Auto pricing is set per million tokens, regardless of which model is used.”

That legacy rate is $1.25 input / $6 output per million tokens, flat, whatever model Auto picks. After the 7th, Auto bills at the list price of whatever it routed to — from Composer 2.5 at $0.50/$2.50 up to Claude Fable 5 at $10/$50. That is a twenty-fold spread hiding behind one setting.

Teams and Enterprise also pay a Cursor Token Rate of $0.25 per million tokens on top of the model’s API price, for any third-party model. Cursor’s own Grok and Composer models are exempt — so the routing decision now has two cost levers, and the cheaper one points at Cursor’s own inference.

If you’re on a negotiated Enterprise order form, check it: Cursor’s docs say the public default applies “unless a negotiated order form says otherwise.” That’s a question for your account rep this week, not next month.

December 31: Gemini Flash doubles

Gemini 3.7 Flash launched August 13 at an introductory rate. Google’s published plan is for that rate to end on December 31, 2026, after which input and output roughly double.

One caveat. Cursor’s model table — which we read directly — shows Gemini 3.7 Flash at $0.75 input / $3.50 output. Secondary reporting on Google’s own intro rate consistently says $0.75 / $3.75, doubling to $1.50 / $7.50. We could not reach Google’s pricing page to confirm the output figure ourselves, so treat that exact number as reported rather than verified. The date and the direction are consistent across every source we checked.

Why a model price belongs on a tool-pricing calendar: if your builder routes to Flash for cheap work — and several do — a doubling at the model layer reaches your bill without anyone announcing a price rise. See which model powers your vibe coding tool.

The one that got cancelled

On July 2 we told readers that Claude Sonnet 5’s introductory $2/$10 rate would rise to $3/$15 on September 1. Anthropic cancelled that on August 10 and made the intro rate permanent. We corrected the original guide yesterday.

It belongs on this calendar as a strike-through rather than a deletion, for two reasons. It’s the only move all year that went in the customer’s favour. And it’s a reminder that scheduled changes get cancelled, which is exactly why a calendar needs a maintenance date on it.

What we could not put a date on

Honest gaps, so you know what this page isn’t covering:

• Lovable, Replit, Bolt and Base44 publish no forward-dated pricing changes at all. Their prices move, sometimes substantially, but they arrive as a live pricing page rather than an announcement with a date. We check these monthly; there is nothing scheduled we can point at.
• OpenAI and Midjourney return 403 to automated reads of their pricing pages. Any figure you see quoted for either is third-party unless the writer says otherwise.
• Regional pricing is now a real fork, not a currency conversion. Cursor launched a Start plan at ₹649/month available only to developers in India, which covers Cursor’s own models but excludes the third-party model pool, Auto, Bugbot and on-demand usage entirely. That’s a different product, not a discount. If you’re outside the US, check the vendor’s page from your own connection.

How to actually use this

Three habits, and they’re all about the allowance rather than the price.

Diary the dates. Put September 7, September 28, October 1 and December 31 in a calendar with a reminder to check your usage dashboard, not your invoice. The invoice arrives a month late. September 28 is the one that needs action before the date rather than after it — that’s the Copilot code review effort setting.

Find your overage default before you need it. For each tool your team pays for, answer one question: when the included usage runs out, does it stop, throttle, or bill? Copilot bills. Cursor bills. Notion pauses. You want to know which one you’re on before the pool empties, not after.

Stop comparing seat prices. A $19 seat with 1,900 pooled credits and overage on by default is not cheaper than a $39 seat with a hard stop. The seat price has become the least informative number on the pricing page, and every change on this calendar is proof.

---

Maintained page. Last verified September 1, 2026. We re-check each vendor’s pricing docs when a listed date passes and add new dated changes as they’re announced. If a date on this page has passed and the row is still here, the check is overdue — tell us.

Changes since publication. September 1: marked the Copilot credit expiry as landed; added September 28 (code review default effort) and October 1 (upfront seat billing), both from GitHub’s own changelog.

One related date that isn’t a price and so isn’t in the table: on September 28 Copilot chat data retention on github.com goes from 28 days to the life of the account, and on December 31 the zero-retention exemption for Claude Fable 5 and 5.1 expires. Both are covered in your AI coding tool just started keeping your data longer.

Sources read directly: Cursor — Models & Pricing, GitHub Docs — usage-based billing for organizations and enterprises, GitHub Changelog — upcoming changes to Copilot policies and billing, Anthropic — Claude Platform pricing.