Auto mode usage increased dramatically on the same project — has routing changed?

Support Help
,
1

Cursor Team,

I am posting this here because I need clarification from Cursor support regarding my account usage.

I have been a paying Cursor user since February. Until my most recent billing cycle, my monthly allocation was generally sufficient for my normal development work.

I am still working on the same application and using a broadly similar workflow, but recently I have seen a dramatic increase in usage consumption.

What previously lasted me an entire month now appears to be consumed in roughly 10 days.

The increase was significant enough that I upgraded to the $60/month Pro+ plan. However, after only about 2 days of normal use, my account is already showing:

  • Auto + Composer: 4%
  • API: 7%

I have never configured OpenAI, Anthropic, or Google API keys, and On-Demand Spending is disabled.

From my perspective, I upgraded specifically because of the increase in consumption, yet I am now concerned that even the higher-tier plan may be exhausted much sooner than expected.

I do not understand:

  • Why API usage is being recorded when I have no personal API keys configured.
  • What exactly counts as API usage under Cursor Pro+.
  • Whether anything has changed recently in usage accounting, model routing, context loading, repository indexing, or quota calculations.
  • Whether the current rate of consumption is expected.

The timing appears to coincide with a recent Cursor IDE update, although I cannot say with certainty that the update is the cause. It is simply when I began noticing the change.

I am not reporting a billing dispute. I am asking for clarification because the change in usage behavior is substantial enough that I can no longer reasonably predict how long my available quota will last.

I would appreciate a response from a Cursor team member.

Thank you.

Additional context for Cursor staff:

I want to emphasize that this is not a question about a specific charge.

My concern is that I have been using Cursor since February, and until my most recent billing cycle my usage was predictable and comfortably fit within my plan.

I am still working on the same application with a broadly similar workflow, yet what previously lasted an entire month now appears to be consumed in roughly 10 days.

This increase was significant enough that I upgraded to the $60/month Pro+ plan. Despite that upgrade, after only about 2 days of normal use my account is already showing:

  • Auto + Composer: 4%
  • API: 7%

I have never configured personal API keys, which makes the API usage particularly confusing.

I am hoping someone from the Cursor team can help explain what changed and whether this level of consumption is expected.

One of several examples that surprised me:

A recent cleanup task consisted of:

  • removing temporary diagnostic logging
  • updating a stale verifier assertion
  • running a small set of verification scripts
  • preparing a suggested commit message

The prompt was narrowly scoped and did not require repository-wide analysis or architecture review.

However, the resulting run consumed approximately 3.3 million tokens in Auto agent mode.

I am not claiming this is necessarily incorrect, but it was unexpected for a relatively small maintenance task and contributes to my concern that token consumption has increased substantially compared to earlier months.

Can Cursor provide guidance on:

  • whether this level of token usage is expected for such a task,
  • whether recent changes have altered how repository context is gathered,
  • and whether there are recommended settings or workflows to reduce unnecessary context expansion during routine maintenance work?

This would help me understand whether the behavior I am seeing is expected or whether something has changed in recent versions of Cursor.

7

A few clarifications on each of your questions.

On the “API: 7%” usage: This does not mean you’re using a personal API key. The “API” category on the usage page tracks requests made through Cursor’s programmatic API features (like Cloud agents). Since you have no personal API keys configured, this likely reflects cloud agent runs or background operations. It’s separate from your own BYOK setup.

On what changed: Auto mode has fixed token rates ($1.25 per 1M input tokens, $6.00 per 1M output tokens) regardless of which model it selects. However, the key driver of higher consumption is agent mode behavior: each time you submit a prompt, the agent can perform many steps behind the scenes (reading files, searching code, running tools, spawning subagents), and each step consumes tokens. A single “simple” cleanup task can legitimately generate dozens of backend calls, which adds up fast.

On the 3.3M tokens for a cleanup task: This is higher than typical but plausible for agent mode, especially if the agent read many files, ran verification scripts, and processed context across your codebase. The agent’s context window grows with each step.

For more predictable usage going forward:

  • Select a specific model instead of Auto (e.g., Composer or a mid-tier model). Auto optimizes for quality, which can mean routing to more capable (and thus token-hungry) models. Selecting a model directly gives you more control.

  • Review your Auto model list in Cursor Settings > Models. Auto may have auto-enabled newer models. You can disable specific models you don’t want Auto to route to.

  • Keep prompts focused. Narrower prompts tend to trigger fewer agent steps.

  • Start new conversations regularly rather than continuing long threads, which accumulate large context windows.

  • Check your usage page to see per-request token breakdowns, which can help identify which tasks consume the most.

Your observation is valid - the same workflow can consume more tokens over time as agent capabilities and model routing evolve. This is something we’ve heard from other users as well and our team is tracking.

Let me know if you have any other questions!