Credits system

The MeinGPT credits system enables fair and transparent billing for AI usage. Whether credits are consumed and how many depends on the plan, the model chosen, the length of the conversation, and the features used.

What are credits?

Credits are the consumption currency inside MeinGPT:

• 1 credit = €0.01 (1 cent)
• Prepaid system — credits are purchased ahead of time and managed at the workspace level
• Workspace-wide — members, teams, assistants, and workflows all draw from the same balance

Credits are managed by workspace admins. As a user you see your share of consumption and possibly your personal usage limit — but typically you don't purchase credits yourself.

How credits are consumed

Credit consumption depends on three factors:

• Model choice — High-priced models (e.g. Claude Opus, GPT-5.6 Sol) consume significantly more per token than cheap models (e.g. GPT-5 Nano, Gemini 3.5 Flash-Lite). The per-token table is on the admin model overview.
• Input and output length — measured in tokens. Long conversations with verbose answers consume more than short questions.
• Feature usage — image generation, translation, audio transcription, and workflow steps have their own prices.

Scheduled tasks always draw on the workspace credit balance, regardless of your contract: the model calls of a scheduled task are charged per token at the selected model's rate, just like a chat, but they are deducted even if your plan includes regular chat usage. There is no fixed fee per run, and tasks that use the free model GPT-5 Nano consume no credits. If the balance runs out, the task pauses until an admin has topped up credits and switched it back on. This usage does not appear in the personal usage overview (sidebar and profile), but it does appear in the usage export.

Example calculation

A chat with GPT-4o:

- Input: 523 tokens × 3.00 € / 1M = 0.16 credits
- Response: 1,247 tokens × 12.00 € / 1M = 1.50 credits
- Total: ~1.7 credits = 0.017 €

Tool calls don't carry their own flat fee - their inputs and outputs count as normal tokens of the surrounding chat response and are billed through the regular model budget.

A normal chat costs a few cents. Image generation, high-resolution models, and complex workflows push consumption up significantly.

Where you see your usage

As a user you find your usage in these places:

• In the sidebar — a small chart with your personal usage limit (if an admin has assigned you a usage tier). The link in the usage details takes you to your profile.
• In your profile — under Settings → Profile you see your consumed budget, limits, and the split across categories (text generation, image generation, API usage, other features, and meetings). If a workspace admin enables the model view, each category also shows costs by model used. It includes only charged credit transactions with a recorded model; usage of free models, usage covered at a flat rate by your plan, and transactions without a model do not appear there. On All-inclusive plans the display never appears. On other plans it appears if at least one personal limit is set for you (including €0) or a workspace admin has turned on Show usage without personal limits under Settings → Usage limits. How much you have already used does not matter. If you do not see the display and your plan is not All-inclusive, ask a workspace admin to turn the option on. Amounts are shown in euros; the platform does not show token counts there.
• In the chat itself — once a single message (from roughly 10 credits) or the whole chat (from roughly 50 credits) gets unusually expensive, a banner appears above the input field with a cost estimate for your next message and a "Switch to a cheaper model?" link.

Two of the places above — the in-chat warning and the per-day cost breakdown — only ever show estimates, not a final accounting:

When you hit your limit, you get a warning at 50 %, 75 %, 95 %, and 100 %. At 100 % other models get disabled for you, but you can still manually select our free model (GPT-5 Nano) and keep working with it — you don't get locked out completely. The warning dialog lets you request an upgrade to the next-higher usage tier. This fallback only applies to the text generation category — image generation, API usage, and the "Other" category for translation and manually started transcriptions have no fallback model and are fully blocked once their limit is reached.

As an admin: who caused the usage?

For any period you choose (for example the last 7 days), you can see costs per user, model and access type (platform or API) in the AI cockpit on the User statistics tab (not available on All-Inclusive plans), and download them as CSV. The chart “Usage over time” under Settings → Usage & Billing shows the trend per day and model. The usage export is aggregated by month and is not suited to day-level comparisons. Usage of individual chats cannot be analysed; narrow down anomalies by user, model and day.

Export: usage is not the same as balance movement

The usage export has four sheets that you combine freely. Each answers a different question:

• Chargeback summary shows charged usage broken down by user, month and model. This is the sheet for internal cost allocation.
• Monthly summary shows one row per month with balances and monthly totals. The Credits Used column contains all AI usage in that month, including usage covered by an inclusive/fair-use quota. Charged to Balance € shows only the part of it that was charged to your balance.
• User × Model matrix shows a pivot table with one row per user and one column per model used, amounts in Euro, for a quick controlling overview.
• Bookkeeping entries shows every balance movement that is not AI usage: top-ups, initial credits, and manual corrections.

Choose Excel (.xlsx) or CSV (.csv) as the file format. A CSV file contains only one sheet; choose Excel to combine several.

The Chargeback summary and User × Model matrix contain only AI usage that was charged to your balance. The total on the Chargeback summary sheet is therefore deliberately not the change in your balance. It leaves out usage that ran against an inclusive quota or a free model and was never charged, and it leaves out every booking that is not AI usage.

To understand why your balance moved by a particular amount over a period, export Chargeback summary and Bookkeeping entries together — only both sheets side by side give you the full picture. If a difference remains after that, contact support and we will reconcile the period transaction by transaction.

Cents vs. euros — important note

In the platform, amounts are sometimes shown in credits (= cents), sometimes in euros:

• Credits are cents. 8,500 credits = 85.00 €.
• Euro amounts read directly.

If a number looks surprisingly large, always check the unit. The platform tries to label the unit explicitly everywhere.

Using credits efficiently

A few tips that matter most in practice:

• Match the model to the task — use cheap models for simple tasks, premium models only where you really need them. The price gap between the cheapest and the most expensive text model is a hundredfold or more (see All models).
• Keep conversations focused — when the topic changes, start a new chat instead of continuing in a long thread. Every new message resends the entire previous history as input; the cost per message grows with the chat's length, and the total for a long chat rises faster than linearly. A new conversation per topic keeps the input small. Prompt caching (see Cached input tokens) and the optional automatic chat summarization soften this but do not remove it.
• Agent runs with large attachments — Each tool step of an agent is a separate model call that again includes the earlier history, attached documents, tool results and previous drafts as input. Consumption therefore grows with the number of steps (Assistant = 5, Agent = 20, Agent MAX = 50) and with the size of the history. You lower it with a cheaper model, a lower effort under "Thinking" and, in chats without an assistant, a lower Tool step limit under Settings > Chat (workspace admins). A lower limit can stop the agent before it has read everything. Alternatively, keep large case-file collections in a data source instead of attaching them; the agent then reads selected passages, but may miss parts.
• Use workflows and assistants for recurring tasks — they reduce manual input and often use cheaper models for individual steps.
• Image generation with care — high-resolution images (GPT Image 1) cost from 0.30 € per image, fast models (FLUX.2 Klein) from 0.018 €.

Frequently asked questions

Do credits expire? No. Purchased credits remain valid indefinitely.

Can I return credits? Already purchased credits can't be returned. For special cases, reach out to our support.

Am I charged if a message fails or gets cut off? For a single request that fails before producing any output, no credits are charged. For larger requests that run in multiple steps – for example reviewing an attached document – steps that already completed are charged even if a later step fails or the request doesn't finish overall. This isn't a double charge; it reflects processing that has already happened.

What do meeting bot recordings cost? Bot recordings cost €0.80 per actual audio hour. Six minutes cost 8 credits or €0.08. Recording, transcription, and the standard summary are included. The charge appears in the separate Meetings category and comes from the shared workspace balance, including on All-Inclusive. The personal meeting limit restricts usage but does not provide additional credits.

What if a meeting transcription fails or is only partial? Bot recordings with no usable transcript are not charged. For a partially delivered recording, only the delivered audio duration counts. Manually uploaded audio and video files and browser recordings keep their existing transcription rules: completely failed transcriptions are not charged; when the service returns a result, the reported audio duration counts. If a result is faulty, contact support with the meeting name and date.

What happens at the personal meeting limit? Once the limit is reached, no further bots start for that person. A recording already in progress is not stopped. Running or simultaneously started recordings can therefore exceed the limit. A refused meeting needs to be scheduled again once the restriction is lifted.

What happens when the workspace balance hits zero or goes negative? Paid models are then blocked for every member until your admin tops up — a negative balance is treated exactly like a balance of €0.00, there's no soft threshold in between. You can still manually select our free model (GPT-5 Nano) and keep working with it during this time; every other model returns the "Workspace has no credits left" error. To prevent this, the admin can set up automatic top-up — but that only applies to paid plans. On a trial workspace, auto-topup stays paused even if configured; the only options there are a manual card purchase or upgrading to a paid plan. Note that the free model only accepts PDFs and images (JPEG, PNG, GIF, WEBP) as chat attachments — Word, Excel, or PowerPoint files, and plain-text files, can't be uploaded to it. Attempting to upload one of these formats fails with the "File type not supported" error.

What about the API? Switching to the free model only helps in the chat. Through the MeinGPT API – and therefore also in tools such as VS Code extensions or coding agents that connect with an API key – every request is rejected with status 402 once the balance is 0 or below, regardless of the model. Some tools show their own message instead of 402 (for example "Quota exceeded").

Buddy is affected too: Unlike the main chat, you can't switch to a free model yourself as an individual member in Buddy – Buddy requests will fail during this time. A workspace admin can, however, set a fixed model for the whole workspace under Settings → Buddy → Buddy model, for example GPT-5 Nano, so Buddy stays usable even with an empty balance.

What happens when only my personal limit is hit? Other models get disabled for you, but you can still manually select our free model and keep working with it — answers will be a bit less capable. The warning dialog lets you request an upgrade to the next-higher tier. Image generation, API usage, "Other", and "Meetings" have no fallback model: once that category's limit is reached, further requests in that category are blocked until the next billing period or a tier upgrade.

How accurate is the forecast? The usage forecast is based on your historical usage and is a guideline — not a guarantee.

For workspace admins

If you manage your workspace's balance: