Cursor Max Mode: Legacy Plans, Cost, and Context

Max Mode lets a supported model read beyond its default context limit. Context is the prompt, selected code, retrieved snippets, rules, tool results, conversation history, and other material sent with a request. A larger window can help when the task crosses many files or when a long session would otherwise discard something important.

It does not make the underlying model smarter. It gives that model room to consider more material at once. That distinction matters because a weak plan, irrelevant files, or a missing source can still produce a bad result inside a large window. Start with good retrieval and explicit scope, then add context only when the task needs it.

Check the billing model before checking the model picker. Older request-based plans can expose Max Mode for supported models. Current usage-based plans meter actual model use and show context selection without the legacy label. A forum screenshot from another plan is not reliable setup guidance for your account.

The capability did not disappear from current plans. The control changed. You can still choose more context where Cursor supports it, but you do that through the context selector rather than a Max Mode switch. This is why current documentation can discuss large context without telling every user to enable Max.

For team documentation, write both paths in one sentence: legacy request-plan users may see Max Mode, while current usage-plan users choose context directly. That keeps an internal guide useful during a migration and prevents support from treating two valid interfaces as a product bug. Record the plan beside any cost example because the same model name does not prove the same billing path.

Use Max Mode only after you confirm that the account is on a legacy request-based plan and that the chosen model supports it. The model picker is the source of truth for that session.

1. 1Open the model selector in the Agent or chat surface you are using.
2. 2Choose a supported model and inspect the context options shown for your plan.
3. 3Enable Max Mode only for the session that needs the larger window.
4. 4Review the usage result after the task before making it a team habit.

On a current usage-based plan, follow the same decision process but choose the context size directly. Do not search for a hidden Max switch or change plans merely to reproduce an older interface. The useful question is how much context the task needs, not which label appears beside it.

Use extended context when the missing information is distributed and cannot be retrieved reliably in smaller pieces. Examples include a cross-layer migration, a refactor with callers across packages, or an investigation whose evidence sits in code, tests, logs, and configuration. A long file by itself is not enough reason.

A simple test is to rerun one representative task with a clean prompt and the smallest context that should work. If the narrow run misses a known dependency and the wider run finds it without extra correction, the larger window solved a real problem. If both pass, keep the smaller choice. If both fail, investigate retrieval, instructions, or model fit before increasing context again.

On a legacy request-based plan, Cursor bills Max Mode at the selected model's API rate plus 20%. The total still depends on actual input, cache, and output tokens, so there is no honest universal price for one Max chat. A long agent run can contain many model calls, each with a different amount of context.

The 20% surcharge is only one part of the difference. Extending context may send far more input on every turn, so token volume can matter more than the multiplier. Check the usage dashboard after a representative task and set team guidance from measured runs rather than from a fixed request conversion.

For a team baseline, capture the model, context choice, reasoning effortA model dial (high / medium / low) for how hard the model thinks before responding, set separately from which model and how fast it runs. Press Enter for the full definition., Fast state, number of agent turns, useful outcome, and final usage. Compare at least a few tasks from the real repository. A one-line policy such as 'use Max for large refactors' is too broad; a measurable trigger such as 'standard context repeatedly omits a required package after explicit selection' is easier to teach and review.

These controls solve different problems. Context changes how much material a model can receive. Reasoning effort changes how much it deliberates. Fast uses priority processing for the same model at a higher rate where offered. Cursor Router chooses a model for Teams and Enterprise requests according to Cost, Balance, or Intelligence. Router Intelligence is not a context-window setting.

Choose each axis independently. A difficult algorithm may need high reasoning with modest context. A broad mechanical migration may need more context with ordinary reasoning. An interactive debugging session may justify Fast. Turning every control up together makes the bill larger without proving a better result.

This separation also makes troubleshooting faster. If the answer ignores a file, inspect context and retrieval. If it sees the facts but cannot solve the logic, change reasoning or model. If the result is good but arrives too slowly, test Fast. If a team wants automatic model choice, tune Router. Changing one axis at a time gives you evidence instead of a bundle of expensive guesses.