The free plan covers everyday messaging — direct messages, groups, channels, voice notes, files, reactions and threads. The AI runs on credits: every plan includes a daily allowance, and paid plans add a monthly pool, stronger models and higher limits.

Which models you get

Standard models — on every plan, including free: Claude Sonnet 5 · GPT-6.1 Sol · Gemini 3.8 Flash

Advanced models — on Pro and Ultra, in addition to all Standard ones: Claude Opus 5.5

You pick the model per chat by typing @ in the composer. The picker marks models above your plan, so you always see what an upgrade adds. You can put more than one model to work in the same conversation — summon a second one to check or extend the first answer. We add and retire models as their providers update them.

Paid plans also raise limits on AI memory, file uploads and voice transcription.

What uses credits

A reply from an AI model, a web search it runs to answer, an image it generates, a document it writes or edits, and the memory it uses to keep track of you.

Never charged: messages to people, voice messages including transcribing them for the AI, and any AI action that does not complete — if a model fails, is unavailable, or refuses a request, nothing is deducted.

Images cost more than replies

Generating an image is the most expensive thing the AI does — a single image can cost more than a whole day of conversation. What it comes to depends on the image model, how many images you get, the size and quality, and how many reference photos go into the request.

When you ask for an image in a chat, the assistant decides on its own which photos from that chat to use as references and whether one image or a short series answers you best. Those choices change the price, so a request can cost more than you expected. Every image keeps a record of what was actually used, and Settings → Usage & billing shows what it consumed.

Who pays for a reply

The person whose message the AI answered. That is you when you summon a model with @, when you reply to something it wrote, and — in chats where the AI answers on its own — when your message is the one it responded to.

In a group, the reply is charged to whoever triggered it, never to the other members and never to the admin who switched the AI on. Enabling the AI in a group does not put its usage on your account.

Chats have a setting for when the AI speaks: only when mentioned, on every message, or on its own when it judges it has something to add. The check of whether it should respond at all is free — you are only charged when a model actually replies.

Where your credits come from

Pool	Where it comes from	Refills	Expires
Daily	Every plan, including free	Every day at 00:00 UTC	At the end of the day — it does not carry over
Monthly	Your paid plan	Every billing cycle	At the end of the cycle
Bonus	Welcome, referral or goodwill credits	No	Only if we state a time limit when granting it
Top-up	Credit packs you buy	No	No

They are spent in that order: daily → monthly → bonus → top-up. Credits that would expire anyway go first, so the ones you paid for are used last.

Daily credits refill at 00:00 UTC, so the exact moment depends on your time zone. Your app shows the next refill in your local time — open Settings and tap the daily row.

What a request costs

There is no fixed price per message. The amount is calculated from the request itself, and these are the things that move it:

Length of the reply. The single biggest factor. Generated text is the most expensive part of any request, so a full draft costs many times a one-liner.
Whether the conversation is active. The most expensive replies are the ones that start something: the first message in a new chat, or the first message after a break. See below.
Attachments. Photos and documents are read in full — one photo costs about as much as three pages of text, and more than the entire conversation around it.
Length of the conversation. Less than people expect. The model reads the last part of the conversation, not its whole history, so a long-running chat does not keep getting more expensive.
Web search. A reply that searches the web reads every source it opens.
Which model answers. Advanced models cost several times more per reply than Standard ones.
Documents. Writing a document is a long answer and is priced like one; editing an existing one costs much less.

Roughly what things cost

What requests of each kind have typically come to in real usage.

	Standard models	Advanced models
Reply while the conversation is active	10–25	20–60
First message in a new chat, or after a break	20–40	110–130
Reply in a group chat	25–35	60–80
Reply that searches the web	20–40	35–60
Creating a document	25–40	110–130
Editing a document	10–20	25–60
Each photo you attach	+5	+12
Each ~10 pages of a document	+50	+125

These are averages, and they are not limits. Each range is drawn from requests of ordinary length — an ordinary conversation, an ordinary answer. Nothing keeps a request inside a range: a very short answer costs less than the bottom of one, while a long answer, a large attachment or an unusually heavy conversation can cost well above the top.

Why some replies cost far more than others

Model providers cache the material a model has recently processed. When a reply can draw on that cache, it costs a fraction of what it otherwise would; when it cannot, the model reads everything from the beginning and you pay full price for it. That single difference is worth several times the cost of a reply — on Advanced models, close to tenfold.

A reply is more likely to be expensive when the model has less to draw on:

The first message in a new chat — there is nothing cached yet.
The first message after a pause. Caches are held by the model providers, for a period they decide and can change; we have no control over it and cannot promise how long yours lasts. The longer the gap, the likelier it is gone — an exchange picked up hours later, or the next day, will often start at full price again.
Switching models in the middle of a conversation — each model has its own cache, so a second opinion means that model is more likely to read the thread from the beginning.
Group chats, for the same reason: different people summon different models, so the context is rebuilt more often.

Replies are likeliest to be cheap while a conversation is actively going — a back-and-forth in one chat with one model is the least expensive way to use the AI, and attachments you have already sent stay cheap for as long as the cache holds.

None of this is something you have to manage. It is simply why one reply may cost several times another, and why your usage history is the place to see what actually happened.

Why the cost is not shown before you send

It depends on how long the answer turns out to be, and that is not known until it is written. We show the exact amount afterwards instead of guessing.

Where to see what you spent

Settings → Usage & billing shows your balance, each pool separately, and a line for every AI action — which model performed it, what kind of action it was, and what it used.

What happens when you run out

The messenger keeps working exactly as before. Only the AI pauses, and you are told before a turn rather than failing mid-conversation. Daily credits refill the next day; upgrading or buying a pack restores it immediately.

Monthly or yearly

Every paid plan is available monthly or yearly, with yearly costing about 20% less. You can switch between them when you change your plan — see Manage your subscription.