Where do I see how much of the limit is left?

In Settings → Usage: the remaining weekly limit, the refresh date and the spending history. When little is left, a bar with the remaining percentage appears above the input.

What happens when the limit runs out?

A request or task run already under way finishes. After that the app stops sending new messages until the limit refreshes, in a messenger the agent answers on the GPT-6 Luna model, and scheduled task runs wait for the limit to refresh.

Can I avoid waiting for the limit to refresh?

Yes: move to a higher plan or reset the limit. A reset can be bought on a paid Pro or Max and used on a paid or gifted one. It applies once 90% of the week is spent, at most 1 time per subscription month, and is non-refundable once applied.

Why is my limit spent when I hardly used the app?

Any work the agent does in your name spends it: scheduled tasks, replies in messengers, helpers. In an organization every member has their own limit, and colleagues' spending does not touch yours. Where it went is in Settings → Usage, under By task and By chat.

Plans and limits

What each plan gives you

There are two plans: Pro and Max. There is no free plan.

Plan	For whom	What is included
Pro	Everyday work	The GPT-6.1 Sol model, scheduled tasks, images
Max	Intensive work	Everything in Pro, the strongest model, Claude Opus 5.5, a weekly limit several times larger, video

Prices and the full comparison are on the Pricing page.

A plan for your staff

There is no separate team plan: an organization buys Pro or Max for as many seats as it needs. Every seat has its own weekly limit and its own images and video by the plan of the seat. On Max some of the seats can be taken as Pro. How to add or remove seats: Payments, invoices and documents.

How the weekly limit works

Every plan carries a weekly limit for model work, with no monthly cap. When it refreshes is shown in Settings → Usage.

There is no fixed message count: spending depends on the length of the chat, the complexity of the task and the tools you connect.

What spends the limit

Any work the agent does spends the limit, not only the messages you write yourself:

chats: every agent turn, including the searching, file reading and service calls inside it;
scheduled tasks: every run. There is no separate run count, runs spend the same weekly limit. A frequent task (replying to every new review, say) can eat most of the week in a couple of days, even while you never open the app;
the agent's replies in messengers and the helpers the agent starts on its own;
images and video: the same weekly limit, plus their own daily and monthly count (the table is on the Limits page);
API calls of some connections: they show as a separate “…: API calls” line under Usage.

While a task is running it holds part of the limit in advance. If the bar above the input says Limit held by tasks, the room frees up once the tasks finish.

Where the limit went is under Usage: By chat, By task, By integration. In an organization there is also By member.

How to spend it more slowly

Clear long chats: every new message is billed together with everything the agent re-reads. Clear the chat when the topic changes.
Move the Speed and depth scale towards Faster for simple tasks: the closer to Smarter, the more a turn costs.
Keep the Auto model: it picks a model for every message, a fast one for simple things and a strong one for hard ones.
Run tasks no more often than the data changes: every extra run spends the limit even when there was nothing to do. A task cannot be scheduled more often than once every 5 min.

When little is left, a bar appears above the input: “…% of your weekly limit remaining”. If what is left is not enough for a chat this long in particular, the bar says so: clearing the chat helps then.

The limit and chat length are different things

The limit is about the plan: how much model work a week includes. It is shared across all your chats and tasks.

Chat length is about the model's window: how much of the conversation it keeps in view in one chat. When a chat grows long, the agent compacts the older part on its own (Compacting context appears), and the chat history stays in place. This has nothing to do with the weekly limit: a full chat does not mean the limit has run out. A long chat does make every turn more expensive, though, so it is worth clearing.

The limit in an organization

Every member of an organization has their own weekly limit, by the plan of their seat, Pro or Max. Colleagues' spending does not touch yours. If your limit went down while you hardly wrote anything, it was spent by work done in your name: your scheduled tasks, your agents' replies in messengers, helpers. Usage shows it under By task and By chat.

When the limit runs out

The limit is checked when a request starts: a request or task run that began before it ran out finishes. Once it is reached, the app stops sending new messages, in a messenger the agent answers on the GPT-6 Luna model, and scheduled task runs wait for the limit to refresh.

What you can do:

wait for the refresh: the date is in the bar above the input and under Usage;
move to a higher plan: from the bar above the input or from the Pricing page;
reset the limit: it clears the week's spend.

A stopped turn does not resume by itself: once the limit refreshes or is lifted, press Continue in the chat.

If you are in an organization

The limit depends on the plan of your seat. Once it is reached, it refreshes on its own; an administrator can bring it back sooner by changing the plan.

Limit reset

Buying a reset works on a Pro or Max bought from us: not on the trial week and not on a plan billed through a marketplace.
Using a reset works on a bought or gifted Pro or Max. A bought reset does not have to be spent at once: it is kept under Settings → Billing → Usage limit resets and waits for the week that needs it.
When it applies: once at least 90% of the week is spent. If the limit is not used up yet, the reset is not spent.
How often: at most 1 time per subscription month.
Refunds: an applied reset is not refunded, an unused one is refunded in full (Refunds).

Trial

An account that has never paid gets Pro for 7 days, once. No card is needed, and the trial does not turn into a paid subscription on its own.

One of three actions starts the week:

the first message to the agent, in the app or to the bot in a messenger;
an answer to the agent's card in the chat;
the button in the trial offer.

In an organization only the person who manages billing starts the week. Other members wait until they start it or pay for a plan.

While the week runs, the organization has 5 seats at no charge: invite colleagues and try the agent together. Trial seats cost nothing and do not renew when the week ends. At checkout you're offered as many seats as members and invites take; you can't pick fewer, so remove anyone extra first under Settings → Members. A marketplace's or an app store's trial has its own seat count; these seats don't apply to it.

When the week ends, without a plan your chats and files stay in place and open for viewing, while sending messages and running tasks are closed until you pay.

Subscription and your own model key

Turn off auto-renewal in Settings → Billing: access stays until the end of the paid period.
Refunds, invoices, certificates and seats: on the Payments, invoices and documents page.
Your own model key is not available to every customer. The Your own models section (a model on your servers or behind a corporate proxy) is turned on by request: write to support. It accepts models in the OpenAI-compatible and Anthropic Messages formats.