DocsOpen Moon AI
Browse the docs

Models and plans

How much of your limit a very high cost model takes on your plan, the lighter model offered, Auto making the rest last, and the free plan.

On this page

The same model doesn’t weigh the same on your limit depending on the workspace’s plan. Moon AI shows you how much of your limit a message takes, offers a lighter model when it should and takes this into account in “Auto” mode. On the free plan, the most costly models are reserved for paid plans, and a conversation open on one of them goes on with an included model. A few models are reserved for Constellation plans and above.

The share of your limit in the model pickerLink to the The share of your limit in the model picker section

In the model picker, “All models” shows three cost bars for each model: “Low cost”, “Medium cost” or “High cost”. A “Very high cost” model for your plan has its bars in red, and the share of your 5-hour limit a typical message uses is written next to them, for example “≈ 25%”. “Whole limit” means that a single message can use your whole 5-hour limit. On hover, the full sentence shows, for example “One message uses about 25% of your 5-hour limit”.

This share depends on the plan of the open workspace: the same model takes a larger share of the limit with Lune than with Univers, and it can be very high cost in one workspace and not in another. It is based on the actual cost of the messages sent to this model over the last 30 days, or on an estimate when it has seen little use yet. Compare the plans on the plans page.

The notice above the message boxLink to the The notice above the message box section

When you choose a very high cost model for your plan, a notice appears above the message box, for example “… is usage-heavy: about 25% of your 5-hour limit per message.” In agent mode, it adds that one request often uses several times more. When only a few messages are left at this pace in your 5-hour session, it adds “At this pace, about … messages before your 5-hour limit.”, or “At this pace, your 5-hour limit is reached with the next message.”

Notice buttonWhat it does
“Use …”Switches to a lighter model of the same level, included in your plan, when there is one.
“Switch to Auto”Offered when there is no lighter model, outside agent mode: Moon AI picks a suitable model for each question.
“Let Auto choose”Shown when you mentioned this model with @ in “Auto” mode: the mention is removed.
“Hide for this model”The cross hides the notice; it only comes back for this model when only a few messages are left before your 5-hour limit.

During a comparison, and for an agent or a workspace model, the notice has no button. It is for information only: you can still send your message. To track your consumption, see Usage limits.

Auto makes what you have left lastLink to the Auto makes what you have left last section

In “Auto” mode, Moon AI also looks at where your usage windows stand: the 5-hour session, the week and the month. When one of them is well used, a model’s price counts for more in its choice, and it sets aside a model whose answer would take too large a share of what you have left, so that the limit doesn’t fall in the middle of an answer. As long as your windows stay far from the limit, Auto chooses as usual.

See Auto mode and the three windows.

Models reserved for paid plansLink to the Models reserved for paid plans section

On the free plan, the most costly models are reserved for paid plans: those with a high price, and those where a single message would take too large a share of your 5-hour limit. The other models stay included. In the model picker, a reserved model shows a padlock and the plan that includes it, for example “Available with Lune”; a click opens the plans window.

In “Auto” mode, Moon AI always chooses among the models included in your plan.

An open conversation goes onLink to the An open conversation goes on section

A conversation open on a reserved model isn’t blocked, and nothing is lost: a model included in the free plan answers instead.

  1. Before you send, a notice above the message box says so, for example “Claude Opus is reserved for paid plans. Your messages go to Claude Haiku.”, with “Use …” to pick that model, and “View plans”.
  2. The model that answers is the strongest one from the same provider included in the free plan, for example Claude Haiku instead of Claude Opus, otherwise a standard Moon AI model.
  3. Under the answer, a line says so, for example “Claude Opus is reserved for paid plans, so Claude Haiku answered.”, with “View plans” under the latest answer.

In a team workspace, a member reads “Ask the workspace owner to change the plan.” instead of “View plans”. An agent or a custom model built on a reserved model isn’t replaced, so that it keeps its instructions: Moon AI answers “This model is not included in your plan. Choose another model.”

Models reserved for Constellation and aboveLink to the Models reserved for Constellation and above section

The “uncensored” versions of some models, tuned to refuse fewer requests than their usual version, are reserved for the Constellation, Galaxie and Univers plans. With another plan, they stay visible in the model picker with a padlock and “Available with Constellation and above”; a click opens the plans window.

In “Auto” mode, Moon AI never picks one of these models for you. In an organization, the plan of your license is the one that counts.

A conversation already started goes onLink to the A conversation already started goes on section

A conversation, a routine or an agent task started on one of these models, for example before a change of plan, isn’t blocked:

  1. Before you send, a card above the message box says so, for example “… is reserved for Constellation plans and above. Your messages go to …”, with “Upgrade to Constellation”.
  2. A model of the same family, without the uncensored tuning, answers instead; if there is none, a standard Moon AI model.
  3. Under the answer, a line says so: “… is reserved for Constellation plans and above, so … answered.”

An agent or a custom model built on one of these models isn’t replaced, so that it keeps its instructions: the request is refused, and “Repair”, on the agent’s page, moves it to a model you have access to. Moon CLI doesn’t offer these models, and a request sent to one of them by Moon CLI, Moon AI for Chrome or an API key is refused: a program never receives an answer from another model without knowing it.

In an iPhone or Android app that doesn’t offer plans, these models say “Available with a paid plan”, and “This model is included in a higher plan.” when you choose one, with no price and no purchase link.

Moon AI

Moon AI in action

0:00

Explore

Screenshots of chat.realmoon.ai taken on October 29, 2026, with fictional data.

Moon AI

Your preferences

Visits counted without cookies.

Our server counts visits without cookies or trackers. Your theme and your answer about ad measurement are remembered on this device.

Ad measurement

With your consent, our ad networks’ trackers (Google) measure our ads, from the visit to the sign-up and then to the first subscription and its amount, and may show you our ads. Your answer is valid for 13 months; you can change it here at any time.

No answer saved on this device.

After you decline, no tag is loaded anymore and the site deletes the cookies they had set on realmoon.ai. Those on the ad networks’ own domains are managed with them or in your browser.

The Moon AI app has its own privacy settings.