Academy
6. Understanding Costs and Credits
Grasp how credits scale with tokens and model weight, and learn the balanced-plus-frontier strategy to control spending.
Every time you send a message in OUPI, the platform counts the tokens exchanged — both what you send (your prompt, attached files, conversation history) and what the model sends back. Your credit consumption equals those tokens multiplied by the model's weight. Frontier models are the heaviest: they can cost ten to thirty times more per token than fast, lightweight models. That means a long back-and-forth with a frontier model is the single most expensive thing you can do in chat. Your balance and each individual exchange show exactly what was consumed, so you always know where your credits went.
The 'balanced-plus-frontier' strategy is the most cost-effective way to work. Start every task with a balanced model — it handles writing, summaries, analysis, and most professional work very well at a fraction of frontier cost. Only escalate to a frontier model when the task is genuinely hard: complex legal reasoning, long multi-document synthesis, or a problem a balanced model has visibly struggled with. For repetitive, short, or low-stakes work, drop down to a fast model. This tiered approach keeps quality high where it matters and avoids burning credits where it doesn't.
Context length also affects cost. The more tokens the model reads — your message, the full conversation history, and any attached files — the more credits you spend. Recent frontier and balanced models can read up to about a million tokens, but feeding them that much is expensive. Two ways to control this: first, use a knowledge base for long or confidential documents so the model reads only the relevant excerpts instead of the entire file. Second, start a fresh conversation when the old history is no longer needed, rather than letting it grow indefinitely.
You can switch models mid-conversation without losing history. A smart move: draft with a balanced model, then escalate to a frontier model only for the one turn that needs deeper reasoning. You pay frontier prices for just that exchange, not the whole conversation.
If you have no strong reason to pick a specific model, use automatic model selection. It reads your request and routes it to a suitable model based on the task, length, and your plan — often choosing a lighter model when a frontier one isn't needed, which saves credits automatically.
Open your subscription page in OUPI. Review your recent credit usage and identify your most expensive exchanges. Check which model was used for each. Ask yourself: could any of those frontier-model exchanges have been handled by a balanced model instead?
Credits scale with tokens × model weight, and frontier models cost 10–30× more than fast ones. The winning strategy: use a balanced model by default, escalate to frontier only for hard turns, and drop to fast for simple tasks. Keep context lean with knowledge bases and fresh conversations. Check your balance regularly to spot where credits go — and let automatic model selection help when you have no reason to choose manually.