Academy
2. What Determines the Cost of an Action
Learn how text tokens, model weight, media generation, agent mission rounds, and sandbox compute time each consume credits differently.
OUPI uses a single currency — credits — to pay for every AI action you take, whether it's a chat message, an image generation, or an agent mission. But not every action costs the same. The price depends on the real resources consumed behind the scenes. There are five main cost drivers:
- Text (chat messages): Cost = number of tokens exchanged × the weight of the model. Frontier (top-tier) models cost far more per token than fast, lightweight ones.
- Media generation: Each image, video, or audio clip has a per-generation price that varies by engine and by the size or duration you request.
- Agent missions: A mission's cost is the sum of all model calls and tool uses across every work round the agent performs.
- Sandbox work: Document generation, code execution, and data processing are billed by compute time.
- Model weight: Heavier, more capable models multiply the cost of every token or call.
Costs are shown before an action when they can be estimated, and always displayed afterwards so you know exactly what you spent.
Text tokens & model weight — the biggest variable
Tokens are the small chunks of text (roughly a few characters each) that AI models read and produce. The longer your conversation, the more tokens are exchanged, and the higher the cost. But the model you use matters even more: a frontier model can cost many times more per token than a fast, everyday model.
This is why long conversations with frontier models are among the heaviest credit consumers on the platform. For routine questions, a lighter model keeps costs low — and OUPI's automatic model selection does exactly that, favoring efficient models when a heavyweight isn't needed.
Media, missions & sandbox — the other cost drivers
Media generation (images, video, audio) is priced per generation. The engine you choose and the output size or duration determine the cost — a short audio clip costs less than a long video.
Agent missions are more complex: the agent works in rounds, calling models and tools at each step. The total cost is the sum of all those calls. Before each round, OUPI checks your balance; if it's insufficient, the mission stops cleanly and keeps everything produced so far — you're never left with nothing.
Sandbox tasks (document creation, code runs, data processing) are billed by compute time. Quick operations are cheap; heavy processing costs more.
Importantly, credits are only debited when an operation succeeds. A failed generation or a crashed sandbox run costs you nothing.
Want to keep costs low? Let automatic model selection choose for you — it picks lighter models for routine work. Write clear, specific briefs for missions (fewer rounds = fewer credits). Store long documents in a knowledge base instead of pasting them into chat. Use temporary chats for throwaway questions.
You can always see what you spent: your credit balance is in the top bar, every mission and media job displays its exact cost afterwards, and the Subscription page shows your plan details and invoices. Check mission costs regularly to learn which actions are expensive.
Open the Missions page in your dashboard. Pick a past mission (or run a quick one with a simple task). After it completes, look at the cost displayed on the mission card. Notice how the number of rounds and the model used affect the total. This builds your intuition for how credits are consumed.
Five things drive the cost of any action on OUPI: (1) the number of text tokens exchanged, (2) the model weight (frontier models cost far more), (3) media engine, size, and duration, (4) the number of mission rounds and tool calls, and (5) sandbox compute time. Credits are only debited on success, and missions stop cleanly if your balance runs short. Use automatic model selection and clear briefs to stay efficient, and check costs on completed missions to learn the patterns.