Grok 4.5 in Cursor: Pricing, Effort, and the Grok 4.6 Upgrade

Cursor introduced Grok 4.5 as its flagship model in July 2026, built with SpaceXAI for coding and other knowledge work. That launch position is historical now. The model remains available, while Cursor's current model page recommends Grok 4.6A Cursor and SpaceXAI frontier model for complex coding and knowledge work; it is a selectable model and is separate from the Grok Bot product. Press Enter for the full definition. for improved instruction following and long-running work. Grok 4.5 still matters for teams comparing behavior or maintaining an approved model policy.

The pool name matters on the invoice. Cursor's models and pricing page puts Grok 4.5 in the Cursor ModelsCursor's plan and usage pool for models it labels as Cursor Models, kept separate from the third-party Other Models pool. Press Enter for the full definition. pool, alongside Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition. and separate from the Other ModelsCursor's pool of supported third-party models from providers such as Anthropic, Google and OpenAI, with separate plan access and usage from Cursor Models. Press Enter for the full definition. pool that third-party requests draw from, and exempts it from the $0.25 per million Cursor Token Rate that Teams and Enterprise accounts pay on third-party work.

The general-model half of that positioning deserves less weight than the coding half. Cursor's docs describe Grok 4.5 as jointly trained for long-running coding and knowledge work. That is a statement of intent, not a benchmark result. Whether the broader knowledge work holds up in your domain is probably something only your own tasks will settle, and the cheapest test is a task where you already know the right answer.

Start with Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition. for routine implementation, and use Grok 4.6A Cursor and SpaceXAI frontier model for complex coding and knowledge work; it is a selectable model and is separate from the Grok Bot product. Press Enter for the full definition. when the task needs more depth. Keep Grok 4.5 when an existing evaluation or policy depends on it, or when you want a controlled comparison with its successor. Cursor's current recommendation points new long-running work to Grok 4.6.

The published rates put a number on that. Composer 2.5The current Composer release, better at long-running tasks and at judging when a job needs a light touch versus deep work. Press Enter for the full definition. bills $0.50 per million input tokens and $2.50 output. Grok 4.5 bills $2 and $6. Composer 2.5 Fast is $3 input and $15 output, while Grok 4.5 Fast is $4 input and $12 output. Base cache reads are $0.20 for ComposerCursor's own fast coding model, tuned for the editor and priced well below frontier models; the recommended day-to-day model for executing a plan. Press Enter for the full definition. and $0.50 for Grok 4.5.

The third row is about measurement, not nostalgia. If Grok 4.5 already clears your acceptance checks, pin both versions to the same representative tasks before changing a team default. Compare useful output, retries, latency and cost. Move to Grok 4.6A Cursor and SpaceXAI frontier model for complex coding and knowledge work; it is a selectable model and is separate from the Grok Bot product. Press Enter for the full definition. when the evidence is better, while keeping a documented fallback during the change.

High, medium and low set how much compute one request gets, on the same model at the same published rate. Cursor's pricing table lists a single rate for Grok 4.5 and another for Grok 4.5 Fast, with no per-effort rows, so the dial reaches your bill through how many tokens a request spends rather than what each one costs.

Which makes a high setting easy to leave switched on. There is no separate line to spot it in, since everything bills per token either way, and the extra arrives as a slightly bigger number against a model you already expected to see. On an individual plan that corrects itself soon enough. Enterprise pools usage across the team, so there it draws from a budget everyone shares.

At five people this does not need a rule. Say out loud which tasks earn high effort, watch the usage dashboard for a billing cycle or two, and correct anyone who is leaving it pinned. Past a few dozen seats that conversation stops reaching everybody, and the lever that scales is the monthly team-wide spending limit Cursor documents for Teams in the dashboard. It caps the month without telling anyone when high effort was the right call, so set the limit and hold the conversation separately.

Cursor Router stops working. The router docs are direct about it. Enabling Cursor Grok 4.5 is a requirement, because the savings depend on having one price-efficient model available for requests that do not warrant a frontier one. On Enterprise the router also honours whatever your team's model access controls allow, and blocking too many models degrades routing quality on its own.

That dependency is easy to trip by accident. An admin trimming the allow-list to an approved shortlist is deciding which vendors the company will send code to, and routing quality is not in frame at that moment. Block something short of Grok 4.5 and the docs say the router substitutes an allowed model instead, and that blocking too many reduces routing quality and can disable the router outright. Reduced quality is close to impossible to notice from inside one thread, so open the allow-list first if Auto starts feeling ordinary. Our Cursor Router page has the modes and the admin controls.

Check the account plan, team model controls, and client version before treating the missing picker entry as a model outage. The launch-era EU restriction has closed, so old regional guidance should not be the first diagnosis.

1. 1Plan. Confirm the account includes Cursor ModelsCursor's plan and usage pool for models it labels as Cursor Models, kept separate from the third-party Other Models pool. Press Enter for the full definition.. Start includes Grok 4.5 at fixed medium effort in non-fast mode; higher plans expose more controls.
2. 2Team policy. Check the team's model access controls and effective organization-group policy.
3. 3Client and session. Update Cursor, start a fresh conversation, and reopen the model picker before escalating with a request ID.

If the model remains missing, the availability fix page walks through the current checks. For a new team standard, compare the same task on Grok 4.6A Cursor and SpaceXAI frontier model for complex coding and knowledge work; it is a selectable model and is separate from the Grok Bot product. Press Enter for the full definition. before spending time restoring a predecessor simply because an older playbook names it.