Grok 4.5 Missing From Your Cursor Model Picker? 3 Gates

The picker only shows models your account can actually run, so a missing model is an eligibility question, not a bug. Work down the three gates. Each has a definitive answer you can verify in under a minute.

That order is cheapest first. You already know your region, your plan is one look at the Cursor dashboard, and team enablement is the only one of the three that needs another person to open a screen you cannot see. Run it backwards and the usual result is a workspace owner asked to switch on a model that was not going to appear in Dublin either way.

1. 1Are you in the EU? Cursor launched Grok 4.5 "in every country where Cursor normally offers models, except the EU," with EU availability planned. If you are, nothing on your side will surface it, so watch the help page for the rollout.
2. 2Are you on a team plan? Team admins must enable Grok 4.5 at cursor.com/dashboard/team-settings/models before members see it. Ask whoever owns the workspace to check the model list there.
3. 3Which plan are you on? Grok 4.5 is included on Pro, Pro+, Ultra, Teams and Enterprise, drawing from 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. usage pool. Free-tier accounts don't get it.

One more check fixes nothing, and I would still run it first, because it tells you which half of the problem you are in. Open the model list somewhere else on the same account. Grok 4.5 ships on desktop, web and cloud agents, the CLI, the SDK, Automations and iOS, and all three gates above are account-level, so a model absent on the web as well as in the editor is an account answer.

Present in one place and missing in the other narrows it to the build in front of you, which is the stale-client case further down.

On Teams and Enterprise there is a fourth setting that can hide Grok 4.5, and it hides every other model along with it. Cursor Router has an Impose Auto control with two strengths: Soft makes Auto the default for each new chat while members can still switch, and Hard locks the model picker to Auto. Both are off by default, so if you are in that state, an admin chose it.

The same feature produces a softer version of this complaint that is not a fault at all, and I suspect it accounts for a fair share of the reports. Cursor keeps the underlying model hidden by default when Auto handles a request in Balance or Intelligence mode, and recommends it that way so a response gets judged on its own merit rather than by the name on it. Somebody can be routed to Grok 4.5 all week and never once see it named.

My first read of a model allowlist was that it settles a spend question. For this model it settles a routing question as well, which changes who belongs in the conversation: whoever owns Auto for the team, not only whoever owns the invoice.

Cursor's router docs are explicit that enabling Grok 4.5 is a requirement for the router to work, since routing needs a cost-efficient model to fall back on when it is not calling a frontier one, and that blocking too many models lowers routing quality and can switch the router off. So turning it off saves the gap between $2 per million input tokens and 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.'s 50 cents, and costs you the thing that was sending everything else to the cheapest model that would do.

Two remaining causes cover almost every leftover case, and they fail in opposite directions. One is your machine being behind. The other is a policy you have no way to see from the picker.

• Stale client. Model rollouts pair with app releases; update Cursor to the current version and restart before digging further.
• Admin model allowlists. Enterprise admins can constrain the model list team-wide. If your org restricts models, Grok 4.5 needs to be added to the allowlist, not just enabled. The same screen at cursor.com/dashboard/team-settings/models covers both.

Enterprise adds a wrinkle to the second one. Model controls sit at two levels, the provider and the individual model, and from a member's seat a provider-level block looks the same as a model nobody switched on. That distinction is an admin's to make, and our model and integration management guide is the better page for that side of it.

If you are the one reporting it, send the admin answers rather than the symptom: the country you work from, the plan the account is on, whether the picker offers other models or only Auto, and whether cursor.com shows the same model list your editor does. Four lines. The last two get left out most often and they are the two that decide whether anybody needs to touch a setting.

If none of this lands, the fastest confirmation of current availability is Cursor's own Grok 4.5 help page. It states regions and plan coverage plainly and is updated as the rollout expands.

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 most of the work. It draws from the same usage pool as Grok 4.5 and carries the same exemption from the Cursor Token Rate on Teams and Enterprise. Input tokens bill at 50 cents per million against Grok 4.5's $2.

Cursor's pricing page lists access to 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. on the free Hobby tier as well, which matters when the plan gate is the one blocking you.

Pinning a frontier third-party model for the hard tasks is the other obvious move, and on a genuinely hard task it is probably the right one. The pricing is not like-for-like, though. Naming a third-party model yourself triggers the Cursor Token Rate on Teams and Enterprise, a flat $0.25 per million tokens on top of the model's own rate, counted on cached tokens as well as fresh ones.

Whether that deserves a policy depends on team size more than on anything else here. With five people, no: someone pins a model on a hard afternoon and the bill barely moves. It gets interesting at the size where pin-whatever-works is the standing habit of a hundred engineers, since by then the fee lands on every cached token in every long thread. Price it before you get there.