Plan limits

How much Grok 4.6 usage do you get on Cursor Pro?

Cursor gives Grok 4.6 no message count. It draws on the Cursor Models pool, shared with Grok 4.7 and Composer 2.5, at $2 per million input tokens and $6 per million output, and the pool resets monthly with your billing cycle.

Checked against the lab's own pages on October 1, 2026. Limits change often, so the date matters.

Cursor doesn't give Grok 4.6 a message count on Pro or any other plan. It draws from the Cursor Models pool, a monthly allowance that's tracked separately from the Other Models pool that Claude, GPT and Gemini use.

That pool is shared, though. Grok 4.7, Grok 4.5 and Composer 2.5 all come out of it too, so a heavy Composer week leaves less Grok 4.6 for the rest of the month.

Pool

Cursor Models, with Grok 4.7, Grok 4.5 and Composer 2.5

Standard

$2/M input tokens, $0.50/M cached input, $6/M output

Fast

$4/M input, $1/M cached, $12/M output

Context window

256k

Resets

monthly with your billing cycle

Which plans include it

Every plan Cursor's usage page names carries the pool Grok 4.6 uses. Start is the odd one out because it has no second pool at all.

On Start (it's India only), Grok 4.6 is fixed at medium effort and standard speed. On Pro and higher, Fast is the default speed tier. Fast costs twice the standard rate per token, so the default setting empties the pool quicker than standard speed does.

Is it worth picking over Grok 4.7?

Cursor's own model page says no. It recommends Grok 4.7 for stronger self-verification and longer-running work, and says Grok 4.6 remains available. What surprised me is that the older model saves you nothing. Both draw on the same pool, and the Grok 4.6 page lists the same $2 and $6 per million standard rates our Grok 4.7 on Cursor page found for the newer model.

Effort is the other lever. Grok 4.6 runs at xhigh, high, medium or low, with high as the default, and higher effort gives it more time on hard tasks (and spends more of your allowance doing it).

What happens when the pool runs out

You'll get a notification in the editor. From there you can turn on on-demand usage, which is pay-as-you-go at the same API rates, or move to a higher plan. Usage resets monthly with your billing cycle and doesn't roll over, and the reset date is on the Spending tab of your dashboard.

Watch Auto and subagents. Cursor's help page says a subagent can run a named third-party model while the picker still shows Grok, and those requests bill Other Models instead.

Cursor doesn't print how many dollars or tokens the Cursor Models pool holds on any plan as of October 1, 2026, so there's no fixed number of Grok 4.6 prompts to quote.