Academy
7. Automatic Model Selection
Use OUPI's automatic model selection to get a good answer without choosing manually, and know when to override it.
OUPI's automatic model selection analyzes your request — its complexity, length, and the resources available on your plan — then routes it to a suitable model without you having to pick one. This means you can start chatting immediately and get a good answer for most everyday tasks: drafting emails, summarizing a report, brainstorming ideas, or answering factual questions. Behind the scenes, OUPI draws from all connected providers (Anthropic, OpenAI, Google, Mistral, DeepSeek, Moonshot, xAI, Perplexity) and maps your request to one of the four model families — frontier, balanced, fast, or specialist — so the right level of power is applied automatically.
Automatic selection is designed to be the sensible default, but it cannot read your mind about every constraint. You should override it and choose a model manually when:
• You need a specific provider for data-sensitivity reasons (e.g., Mistral for French/EU hosting, or a provider your admin has approved). • You are working with a very long document and need to be sure the model's context window can handle it. • You already know the task is hard — legal reasoning, multi-step calculations, complex code — and want to go straight to a frontier or reasoning model. • You want to compare answers from two different models on the same question.
In short: let automatic selection handle the routine, and step in when you have a reason.
If automatic selection gives you a weak answer, don't restart the conversation. Switch to a stronger model (e.g., a frontier or reasoning model) mid-conversation — your full history is preserved, and the new model picks up right where you left off. This 'escalation' trick saves credits on the easy turns.
Understanding the four families helps you override wisely:
• Balanced models are the workhorse — great quality at moderate cost, ideal for most professional work. • Frontier models handle the hardest tasks: deep reasoning, long multi-document synthesis, complex agent work. • Fast models are cheap and quick, perfect for short, repetitive, or low-stakes tasks. • Specialists shine in narrow areas: reasoning models for multi-step logic, code models for generation, and live-web models for current information.
Automatic selection navigates these families for you, but knowing them lets you make a confident manual choice when needed.
Keep an eye on your credit balance. Frontier models can consume ten to thirty times more credits per token than fast models. Letting automatic selection handle routine work usually keeps costs reasonable, but if you manually pick a frontier model for a long conversation, costs add up quickly.
Open the OUPI chat interface. Leave automatic model selection on and send a simple question (e.g., 'Summarize the benefits of remote work'). Then, for the next message, manually switch to a different model family — try a frontier model — and ask a harder follow-up. Notice how the history is preserved and compare the depth of the answers.
Automatic model selection is your best starting point: it reads your request and picks a suitable model across all providers and families. Override it when you have a specific reason — data sensitivity, very long documents, known-hard tasks, or model comparison. You can always escalate mid-conversation without losing history. Remember the four families (frontier, balanced, fast, specialist) and five criteria (difficulty, length, modality, sensitivity, budget) to make confident manual choices when needed.