Task-Based Token Usage Estimation & Model Recommendation

Ideas Feature Requests
1

Feature request for product/service

Cursor IDE

Feature Idea: Task-Based Token Usage Estimation & Model Recommendation

One feature that could significantly improve the development experience is pre-execution token usage estimation combined with intelligent model recommendations for each task.

Before a task is executed, the IDE could analyze its complexity, context size, attached files, and expected output length to provide an estimated token consumption range. Alongside this estimate, the IDE could recommend the most appropriate model based on factors such as reasoning complexity, speed, cost, and expected quality of results.

This estimate becomes even more valuable when correlated with the user’s remaining token quota or usage limits. Instead of showing only raw token balances, the IDE could present meaningful insights such as:

  • Estimated tokens required for the current task.
  • Percentage of the user’s remaining quota that the task is likely to consume.
  • Confidence range for the estimate (e.g., low, medium, or high variance).
  • Recommended model with an explanation of why it is the best fit.
  • Alternative models that trade off quality, speed, or token efficiency.

For example, if a user has 250k tokens remaining and the current task is estimated to consume approximately 35k tokens, the IDE could communicate that the task will use around 14% of the remaining quota. If a more token-efficient model can achieve similar results, it could proactively recommend switching to that model and display the expected savings.

This feature would improve the overall developer experience by making token consumption predictable rather than reactive. Users could confidently plan larger workflows without worrying about unexpectedly exhausting their token budget midway through implementation. It also reduces the likelihood of long-running development sessions coming to an abrupt halt due to depleted tokens, allowing users to focus on building rather than constantly monitoring usage.

In the future, this concept could be extended with additional capabilities such as:

  • Dynamic token forecasts as the prompt evolves.
  • Project-level token budgeting across multiple tasks.
  • Warnings before expensive operations.
  • Automatic model switching based on the user’s remaining quota and task complexity.
  • Historical accuracy metrics comparing estimated versus actual token consumption to continuously improve estimation quality.
5

Hey, thanks for the detailed feature request. I’ll break it down.

Model recommendation / auto-switching: This partly already exists as Auto. Pick Auto in the model picker, and Cursor will choose a model for your request. One key detail is that the advanced version, Cursor Router with Cost / Balance / Intelligence modes, is currently only available on Teams and Enterprise plans. It’s not available on individual plans yet. More about Router: Cursor Router | Cursor Docs

Quota visibility: In chat, there’s a Usage Summary in chat/agent preferences that shows usage during the session. One note is that in the new Agents Window it’s currently hidden if chat status bar is enabled. As a temporary workaround, you can turn off Status Bar entirely in the View menu, and Usage Summary will show up again. Total usage is always visible in Settings > Plan & Usage and on the dashboard at cursor.com/dashboard.

Pre-execution token estimate: There’s no dedicated feature for this yet, like estimating cost before running, percent of remaining quota, or a confidence range. There’s a technical reason, since real usage depends on the model output, tool calls, and extra context the agent pulls in while it runs. That makes an exact estimate hard, and you can only do a rough guess. I’ll pass this on to the team as feedback. A similar request is already being discussed here: Pre-execution cost estimate / confirmation for expensive operations Feel free to add more details there.

Let me know if any of this covers part of what you need.