Don't Kill Auto From the Email Alone — Cursor's Aug 24 Per-Model Pricing

Effective 24 August 2026, Cursor Auto no longer draws against a single flat Auto rate. Usage is priced from the list price of the model each request is routed to. In the same notice, Cursor raised included limits for Cursor Models (including Auto) and said routing behavior is unchanged.

Core takeaway: Auto did not “break.” The subsidy-shaped flat rate ended, the included pool got larger, and your invoice now shows the model rates Auto was already choosing. Check Spending before you rip Auto out of the picker.

Why it matters

Agent loops, cloud agents, Subscriptions, and /goal turn Auto from a convenience toggle into always-on infrastructure. The first public trail for this change was not a polished changelog post — it was a Team email. Feeds filled with “Auto got expensive” and “pin a model forever” before help-center wording fully caught up.

The operational risk is misreading the invoice semantics. A row that shows a model’s list-price equivalent can still be included usage, not on-demand. The opposite error is just as bad: treating “limits went up” as permission to leave Intelligence-mode cloud agents unsupervised for days. If you treat the email as a product outage, you skip the only work that matters — Spending review, default Optimize For policy, and pinning frontier models for long-running goals.

How it compares

Read Optimize For roughly like this:

• Cost — bias toward token spend efficiency (legacy-Auto-shaped axis).
• Balance — default-ish blend of quality, speed, and cost.
• Intelligence — steer harder work toward more capable models.

Current help text says all Auto modes bill at the routed model’s list price, and third-party models also incur the Cursor Token Rate. Docs snippets also mention an Enterprise Auto Cost legacy flat window through 7 September 2026 ($1.25/1M input & cache write, $0.25/1M cache read, $6.00/1M output). Treat that as a plan-specific carve-out: cross-check the email’s general rule against your Billing rows and the live docs for your plan.

Code / config example

The UI is source of truth. Still, write the checklist down so the next email panic does not become tribal knowledge.

# Team ops checklist (Aug 24+)
1. Settings → Billing / Spending
   - Cursor Models pool remaining
   - Other Models pool remaining
   - on-demand enabled?
2. Model picker → Auto → Optimize For
   - default for chat/tab: Balance or Cost
   - default for long agent /goal: pin a frontier model OR choose Intelligence deliberately
3. Optional: show routed model while debugging cost spikes
4. Do NOT treat a list-price row inside included usage as "extra invoice charge"

If your team uses Router via API/SDK, docs describe model id auto-smart with optimize_for of cost | balanced | intelligence. Confirm availability with Cursor.models.list() before hard-coding.

// Illustrative — confirm against current SDK docs for your plan
const result = await cursor.agent.run({
  model: "auto-smart",
  optimize_for: "balanced", // cost | balanced | intelligence
  prompt: "Summarize failing CI and propose a minimal fix",
});

The first time I opened Spending after the email, I nearly flipped Auto off on the “higher rate” line alone. The same screen showed remaining included pool and the on-demand toggle — measuring a week looked cheaper than guessing from the subject line.

Practical use

1. Open Billing/Spending after 24 August. The email says limits go up; it does not publish the new numbers.
2. Split modes by task shape. Short Q&A and tab completion → Balance/Cost. Multi-hour /goal or PR babysitting → pin a frontier model or choose Intelligence on purpose.
3. Watch both pools. Cursor Models and Other Models are separate monthly buckets. When Auto lands on a third-party model, Token Rate can stack.
4. Write a one-line team default. Example: “Default Auto=Balance; long agent jobs pin a named model.”
5. Re-budget Subscriptions and cloud agents. Event-woken agents will chew the pool faster than flat-era intuition expects.

Senior-engineer perspective

This is the familiar AI-coding-tool arc: flat or subsidized pricing builds the habit; agentic volume forces metering to show its face. It sits in the same ops family as Copilot credit clocks and model-deprecation calendars.

The failure mode is rewriting policy from rumor percentages. Cursor said most requests draw at a higher rate than the old flat — not that your repo’s Intelligence /goal empties the pool in three days. Order of operations: sample Spending for a week → set the default mode → pin only the long jobs.

Also: killing Auto and always picking the top frontier model buys predictability and can still overpay for work Router would have handled cheaply. The opposite — trusting Auto while hiding the routed model — makes cost debugging impossible. Showing the routed model when you investigate a spike is a craft tool, not a vanity setting.

Using it in Cursor

• For cloud agents, Subscriptions, and /goal threads, write the model / Optimize For on the task card. “Just use Auto” is now a clearer billing unit.
• When spend spikes, do not start by rewriting the quality prompt. Check Spending for which pool and which list price moved.
• If an agent drafts team policy, put “Spending tab checked” in acceptance criteria. Email-only pricing changes leave documentation gaps.
• Relative to earlier Router, token-efficiency, and Subscriptions drafts: routing is the quality/cost lever; Aug 24 is how that lever prints on the bill.

Looking ahead, team default mode plus “pin for long agents” will stop being taste and start being a monthly ops runbook line.

FAQ

Q. Should I turn Auto off immediately because it got more expensive?
A. Not necessarily. Included limits increased and routing is unchanged. Decide after you see pool remaining, on-demand state, and actual model-rate rows in Spending.

Q. Are Cost / Balance / Intelligence separate SKUs?
A. They are Optimize For modes under Auto. Current help describes billing at the routed model’s list price; verify any Enterprise Cost legacy window for your plan in the live docs.

Q. If I see a model list price inside included usage, is that an extra charge?
A. Often it is accounting against the included pool. Distinguish included draw vs on-demand on the Spending/invoice views.

Q. Is there an official changelog URL for Aug 24?
A. The primary trail is widely reported as a Team email. Cross-check Pricing, Router, and Usage limits help pages plus your Billing UI.

Sources

• Cursor Auto Pricing Aug 24: More Usage, Higher Rate — explainx.ai (Team email summary)
• All roads lead to metered pricing — Arkadiy Kondrashov
• Pricing — Cursor Help
• Cursor Router — Cursor Help
• Cursor Router — Cursor Docs
• Usage and limits — Cursor Help
• Introducing Cursor Router · Cursor