Knowledge

Why the numbers don't match

The model selector, which opens from the model's name at the top of a chat, shows a Price with the caption Avg. stars per message. Yet a message can cost a different amount, and your balance can dip while an answer is still being written. Both are normal. Lexika charges for what the model actually reads and writes, in two steps.

In a workspace on Owner Credit, the owner pays. Every hold, refusal and cut-off below then depends on the owner's stars, and the charges land in the owner's usage list. The Knowledge Base article on Owner Credit and Individual Credit covers who pays.

What the Price figure means

The Price sits along the bottom of the model selector, for the model you point at or else your selected one (on a phone, as a star chip with the number). It also shows in the dialog behind a card's (i) Info button. Eco, Balanced, Premium and models marked Not in your plan show none.

Lexika sets this one number for each model. It is not the price of your next message, which can cost well under or over it. The figure does two jobs.

• A minimum. When your spendable stars fall below it, the model's card shows an amber badge with the stars it needs. If that model is selected, a strip above the composer says so and the send button stays off. The strip's Get stars button opens Settings > Subscription. In an Owner Credit workspace you don't own, the strip says instead that the workspace's plan doesn't cover the model, even when the owner's stars are the cause, and it offers no button.
• A buffer. Lexika adds it to every hold and gives it back when the answer ends. Lexika never charges it.

A hold first, then the real charge

When you send a message that asks the AI, Lexika gathers what the model will read: your Chat instructions, any Brain material, the summary of a compacted conversation, the recent messages and attached files.

1. The hold. Before the model starts, Lexika sets aside enough stars for that reading, plus the Price.
2. The charge. When the answer ends, Lexika swaps the hold for the real cost, which is what the model read plus what it wrote at that model's rates. It can be more or less than the hold.

Held stars aren't spendable, so mid-answer your balance can look lower than it will end up. Each running answer holds its own stars, so a send can be refused until the others finish.

If your stars cover the Price but not the whole hold, Lexika starts no answer. A red notice under your message reads Message failed to send, then You don't have enough stars. If Eco, Balanced or Premium picked the model, Lexika has already taken the routing charge, and it stays.

If a send fails after Lexika sets stars aside but before an answer starts, those stars stay unspendable until 15 minutes after the hold began, then come back without a charge.

What makes one message cost more

Each time you ask, you pay for the model to read the whole conversation again, so messages cost more as it grows. Lexika eventually condenses the oldest part into a summary, which slows that growth. The Knowledge Base article on context compaction explains it.

• Steps. Each web search or tool step mid-answer reads the conversation again, and each one counts.
• Reasoning. Reasoning counts as writing, so a higher Reasoning effort usually costs more. On models that offer it, its button on the composer's control row shows the current level, such as Medium. On a phone it sits behind the three-dot Tools button.

When your stars run out mid-answer

The hold does not cap the charge. Your balance does. When you send, Lexika works out how long an answer your stars can pay for once the reading is paid, and it stops the answer there. Nothing on screen says why. The answer just ends, sometimes looking finished and sometimes like one you stopped. A charge never takes your balance below zero.

If answers end mid-sentence on a low balance, add stars in Settings > Subscription, or start a new conversation, which has less to read.

Stopped, failed and retried answers

What happened	What you see	What is charged
You pressed stop	The text ends where it was	The full reading, plus what was written
It failed	Faded text, and This response failed to generate in a red box or on the red warning by the model's name	The full reading, plus what was written
The send was refused	Message failed to send under your message	Nothing, except a routing charge from Eco, Balanced or Premium

While Lexika writes, the send button becomes a square stop button. In an existing chat, typing in the box turns it back into send. Stopping right after you send still costs the full reading, and so does an answer that comes back empty.

Message failed to send means nothing arrived, so Retry, the red circular arrow beside your message, is safe once you fix the cause. Delivery uncertain means it may have arrived, so Retry first asks Retry this message? Retrying a message that did arrive makes a second charged answer. A delivered message usually clears its notice and starts its answer on its own, so wait a moment first.

Regenerate under an answer opens a model list with Retry with same model first, and Lexika charges the new answer in full at the chosen model's rates. Lexika removes the old answer before it takes the new hold. If your stars can't cover that hold, no new answer arrives, and the old answer and its question are gone.

Other charges one message can add

• Web searches. A model that can use tools decides for itself whether to search, and a chat has no switch to stop it. Lexika charges each search separately.
• Perplexity request fees. Perplexity models add a fee per request when the answer finishes.
• Pictures. Lexika charges separately for a picture from its picture tool, and nothing for one that fails, is refused or is stopped. A model that makes images by itself counts them as part of its writing.
• Eco, Balanced and Premium. When the model name at the top of the chat shows one of these, Lexika picks a model for each message and adds a routing charge. Lexika uses Balanced until you choose a model, so pick one by name to skip the charge. The Knowledge Base article on Eco, Balanced and Premium covers it.

Compacting a conversation costs you nothing. In an existing chat, where your workspace's plan allows it, click Asking AI in the composer (a chat-bubble icon on a phone) to switch to Chatting. Messages there reach no model and cost nothing, unless you @mention the AI.

Seeing what each message cost

Open Settings > Usage, or click the star ring beside the settings gear at the bottom of the sidebar and choose View usage details. The list opens on Last 7 days, and its range menu reaches back to Last 3 months. Individually lists each charge with its date and time and its Stars used, to at most two decimal places. Totally adds them up by Feature.

• A chat answer appears under the model's name as Lexika records it, which can differ from the name in the model selector.
• A picture from the picture tool appears as gpt-image-2.5-sunburst.
• Web searches appear as WEB_SEARCH, the routing charge as AUTO_MODEL_ROUTING, and Perplexity fees as PERPLEXITY_REQUEST_FEES.
• One message can produce several rows, and no row names its conversation.

The sidebar and Settings round your balance down to whole stars, so a charge under one star may not change it.

Cycle stars and extra stars

Settings > Subscription splits your balance into Cycle stars and Extra stars, and Lexika spends cycle stars first. Every month, even on a yearly plan, Lexika refills cycle stars to your plan's allowance, and unused ones don't carry over. Resets in shows the days left. Extra stars, from star packages or from vouchers that add stars, never expire.

Moving up to a bigger plan on the web or through Google Play while the current one runs turns your remaining cycle stars into extra stars. With an App Store subscription, a plan change replaces them instead.

Keeping costs predictable

• In a long conversation, keep a margin of stars above the Price, so sends go through and answers aren't cut short.
• On a low balance, copy what you need from an answer before you regenerate it.