Skip to content

Your Cursor bill has four tanks. They do not mix.

Live Cursor Ultra bill (Aug 18–Sep 18 cycle) shows four separate usage tanks that do not mix: Cursor Models (monthly), Other Models (monthly), Grok Bot (weekly), and Grok CLI (weekly). Display rows are not quota. Estimates and hypotheses are labeled.

August 27, 2026Cursor · Grok Bot · billing · usage

You keep the human work. We build the rest.

By James Brady · August 27, 2026

Four tanks. They do not mix. — ON‑SCREEN

What is on the screen (ON‑SCREEN)

  • Account and cycle: Live Cursor Ultra bill for james@jamesbrady.org, invoice Aug 18 $0.00 Paid, cycle Aug 18–Sep 18. Subscription is SuperGrok Heavy (includes Ultra) with on‑demand $0 / $200 remaining at this snapshot. Four usage tanks are rendered separately and do not mix. The four tanks printed:

  • Cursor Models — shows 82M tokens used, meter marked 0.0% of monthly.

  • Other Models1.6B tokens, 8.5% of monthly.

  • Grok Bot14% used, a weekly bar that resets Sep 1.

  • Grok CLI (Heavy)~3% used, a weekly bar whose size is not published.

The percents are render order only — they are not tank sizes and should not be ranked against each other.

The promo and what it did not do (ON‑SCREEN)

  • The SuperGrok promotion ended Aug 21. New Heavy subscriptions after that date do not get Cursor Ultra.
  • A Grok Bot link is a usage grant for the weekly Bot bar. It does not unlock IDE access, does not create an Ultra plan, and does not fill the Other Models tank. Source: staff confirmation (Colin, Aug 21).
  • “Grandfathering” means: keep Heavy active to keep the linked benefit. Team SuperGrok does not qualify. There is no sunset printed; Sep 18 is the cycle boundary for this bill, not a program end date.

Display rows are not quota (ON‑SCREEN)

  • Row labels (sand-default, sand-automation, etc.) display Bot usage, not the Other Models quota. Weekly Bot is separate from monthly Other Models.
  • Rows observed on the bill: 1.1B default, 157.4M automation, 13.2M sand‑cua (James/CoS attribution), Opus 138.2M, GPT‑5 147.9M. Invoice still $0 at this snapshot.

What “Other Models” mostly is — ON‑SCREEN

Tokens are not equal (ESTIMATE)

Relative mix pricing (USD per 1M tokens). These are estimates for mix‑B, not a bill:

  • Flash: ~$0.933
  • Grok 4.6: ~$3.05
  • GPT‑5: ~$3.76
  • Fast Grok: ~$6.10
  • Opus: ~$10.55

Multiplying list by tokens is not the bill (HYPOTHESIS)

  • Report table totals like $5,314 / $5,967 / $1,469 are not the invoice.
  • A $400 “Other Models” ladder is a HYPOTHESIS (old 20×$20 mental model). It is not on this bill and is not printed by Cursor.

How big is the “Other Models” tank? (ESTIMATE)

  • From the screen: 1.6B tokens at 8.5% of monthly ⇒ 1.6 ÷ 0.085 ≈ 18.82B tokens capacity. This is math on a live display, not an official cap. Do not invert Bot: 1.2706B ÷ 0.14 would be apples to a different tank.

Estimated Other Models capacity — ESTIMATE

Cursor Models lower bound (ESTIMATE)

  • If 0.0% prints below 0.05%, then 82M tokens on the Cursor Models bar implies > 164B monthly capacity. This is a bound, not a printed cap.

On‑demand $200 and what it could buy (ASSUMED)

  • If on‑demand were charged at the assumed 8k/1.5k/20k cache turn, $200 would buy roughly ~2.3k Opus to ~29.6k Flash requests, or ~5.7k Grok 4.6. These are ASSUMPTIONS to make the bar legible, not prices from the bill.

Read this as an operator

The buckets do not mix. Display is not quota. Treat the bill like a dashboard: take the numbers that are printed (“ON‑SCREEN”), label every inference (“ESTIMATE”), and mark any old ladder you are still carrying around as what it is (“HYPOTHESIS”).

Name the Work · Book a Call

Contributors

  • James Brady

    Author · Publish decision