Cursor Auto Pricing: Per-Model Rates and Higher Limits on August 24

Included limits go up. Auto stops being a flat rate. Routing stays the same. It hits on August 24, on the cycle you are already in.

That is the whole Cursor Team email from team@mail.cursor.com, effective August 24, 2026. Thread takes that turned it into "Auto is now Cursor-only," "this is just a price hike," or "they shipped a Grok 4.6 harness" are adding claims the email does not make.

The honest read is narrower and less comfortable: you get a bigger included pool for Cursor Models, including Cursor Auto, and Auto may burn that pool faster because most routed requests will draw at a higher rate than today's single Auto price.

TL;DR — what people are asking

What the email actually says

Cursor's three bullets, condensed:

1. Updated Auto pricing. Auto will use per-model pricing — the rate of the model the request was routed to — instead of one flat Auto rate. Cursor's stated reason: keep serving the most cost-efficient models for everyday work. Cursor also says that for most requests, this means a higher rate than the current flat rate.
2. Increased usage limits. Included usage limits for Cursor Models, including Auto, go up. Applies to the current billing cycle.
3. No change to how Auto works. Auto still routes each request to the model best suited for the task. The models Auto routes to are not changing. Cursor still calls Auto the best option for intelligence per dollar.

An X recap the evening of August 19, 2026 (NIK, about 41.4K views) matched those three points and added two extras the email does not spell out: more included usage on all Cursor Models, and Cursor Ultra via SuperGrok Heavy included too. Use the email for policy. Use the X recap as a rumor-check against your Billing page after August 24.

Is this more usage, or a hidden price hike?

It is both mechanisms at once. Anyone selling only one of those sentences is incomplete.

The limit increase is real as a product statement. Cursor has done this before: July 16 doubled included first-party usage, then July 21 restated the 2× pool without a second multiplier. August 24 is another included-capacity bump, not a restatement of that 2×.

The rate change is also real. Cursor wrote that Auto's new per-request rate is usually higher than today's flat Auto rate. On a subscription, that is how you get a hidden hike without a new monthly sticker: the same agent loop can consume more of the included allotment even after the allotment grows.

What the email does not give you: the old flat Auto rate, the new per-model table, or the new included-token or included-dollar numbers by plan. Until those land on Settings → Billing / Usage, you cannot do the arithmetic of "is my net Auto budget up or down."

Practical check after August 24:

1. Note first-party vs API remaining before heavy Auto sessions (same split we used for the July 2× pool).
2. Run a routine edit and a hard-reasoning prompt on Auto. Compare Usage-page cost or tokens per request against last week's Auto rows.
3. If Auto now lands on a frontier model for work you used to get at the cheap flat rate, treat that as expected — not a bug.

Does "higher rate" mean dollars, or just a faster bite of the pool?

This is the Ed Zitron-shaped question, and it is the right one for subscription coding tools.

Cursor's email talks about Auto pricing and a higher rate, not a new public list price for the Auto product. For most individual and Teams seats, Auto is paid as draw against included usage, then on-demand if you exceed the pool — the same semantics we covered when Cursor restored dollar costs on the Usage page. A row can show a model-rate dollar equivalent and still be included on the invoice.

So "higher rate" in this note most naturally means each Auto request is metered as if you had picked that routed model, which eats more of your allocation when the routed model is expensive. It is not, on the email's text, a separate "Auto surcharge" you pay on top of the model.

If Cursor later publishes an Auto API tariff, that would be a dollar price. They have not done that in this email.

The Usage-page fight in early August is why this wording will confuse people: tokens, included, and estimated dollars are three different columns. Per-model Auto makes the estimated-dollar column more important, not less.

Does Auto now only route to Cursor-native models?

No.

One reply treated "Cursor Models, including Auto" and "cost-efficient models" as proof that Auto would only hit Composer / Grok / other first-party SKUs. The email contradicts that in the third bullet: the models Auto routes to are not changing.

That is the same architecture as Cursor Router from July 22: per-request analysis, Intelligence / Balance / Cost bias, frontier models when the task needs them. Router's launch already billed some Auto modes at the routed model's rate. August 24 generalizes that idea away from a remaining flat Auto rate, which Cursor says was too cheap to keep serving the cheap everyday models.

If your Auto sessions still land on Claude, GPT, or Grok after August 24, that is consistent with the email. If they suddenly cannot, that would be a later product change, not this note.

Did Fable "do this the other day" because of GPUs?

Treat the Fable 5 GPU-capacity joke as a joke, not a sourcing claim.

explainx.ai's Fable coverage is about export-control restore, classifiers, and usage-limit / credits policy — see Anthropic's official Fable 5 redeploy and the July 12 usage-credits cliff. Those posts document quota and access rules, not a Cursor-style Auto flat-rate rewrite, and they do not establish a GPU-shortage number you can paste onto Cursor.

Cursor's stated motive here is economic, not hardware: keep cheap everyday models in the Auto mix by stopping a single underpriced Auto rate. That is the same retention pattern as why labs push agent usage: included capacity and metering rules move more often than the $20 seat.

Has Cursor added a Grok 4.6 harness?

Not in this email.

Grok 4.6 launched August 12 with Cursor and Grok Build on day one, the same $2 / $6 list as 4.5, and a first-week 2× included-usage promo in those two surfaces. That is a model landing in an existing IDE and in Grok Build, not a new Cursor harness announced on August 19–20.

The X recap's "Ultra via SuperGrok Heavy included too" line is about which Cursor / Grok entitlements get the bigger included pool, not about a new runtime. SuperGrok Heavy and Cursor Ultra already appear together as access gates for products like Grok Bot. Do not collapse those plan names into "Cursor shipped SuperGrok inside Auto."

If you want a harness comparison, use existing explainx.ai guides — what an agent harness is and Claude Code vs Cursor — not this billing note.

Same week as Cursor hosting — is this one story?

Two stories, one calendar.

Origin is an in-editor code hosting beta for paid users, announced August 17–18 during a GitHub outage. This Auto email is a metering and included-usage change effective August 24. Origin does not change Auto routing; the Router post already noted that after Origin shipped.

Do not wait for Origin onboarding to "unlock" the usage bump. Cursor says there is nothing you need to do.

What to do before August 24

1. Screenshot or export Usage for a typical Auto week (model column, included vs on-demand). You will want a before/after once per-model Auto lands.
2. If you use Auto as "cheap default," assume routine work may still be cheap and hard work will look more like picking that frontier model yourself. That is the point of dropping the flat rate.
3. If you already pick models by hand, this email is mostly about Auto users and the included Cursor Model pool.
4. After August 24, confirm the new included number on the Billing page rather than trusting recap posts for the integer.

Related on explainx.ai

• Cursor Router — Auto model selection (July 22)
• Cursor doubled usage limits — July 21 clarified
• Limit reset day — Claude, Codex, Cursor (July 16)
• Cursor restores dollar costs on the Usage page
• Grok 4.6 launch — evals and Cursor access
• Cursor Origin — code hosting, August 17–18
• Fable 5 redeploy and usage limits
• Why AI companies want you using agents
• Cursor Auto · Per-model pricing · SuperGrok Heavy · Cursor Ultra

Primary sources: Cursor Team email from team@mail.cursor.com, effective August 24, 2026 · X summary post (NIK, ~41.4K views, August 19, 2026 evening)

Pricing, included amounts, and routed-model lists are accurate only as stated in Cursor's August 2026 Team email and the cited explainx.ai posts. Cursor did not publish the new numeric limits or the per-model Auto tariff in that email — verify Settings → Billing after August 24, 2026. Follow @explainx_ai for updates.