Academy
5. Team and Plan Restrictions on Model Eligibility
Understand how your plan and your administrator's team settings define which models automatic selection can choose from, keeping data with approved providers.
Automatic model selection doesn't pick from every model on the platform — it picks from the models your plan and team settings allow. Your subscription plan determines a baseline set of eligible models (for example, some plans may not include frontier-tier models). On top of that, your organization's administrator can further restrict which providers and models your team is permitted to use. The automatic selector only ever chooses among this filtered pool. This layered approach means that even when OUPI is routing your messages for you, data never flows to a provider your organization hasn't approved. The routing engine still scores each eligible model on task fitness, cost, and reliability, but it simply cannot see or select a model outside your allowed list.
Why do these restrictions matter? Data sensitivity is one of the five key criteria for choosing a model. Some organizations require that content stays with European providers — for instance, Mistral is French and hosted in Europe. Others may approve only specific providers after a security review. By restricting the eligible model pool at the team level, an administrator ensures compliance without relying on every user to remember the policy. Automatic selection respects these boundaries transparently: you still get intelligent routing across task types, complexity levels, and cost tiers, but only within the approved perimeter.
Even within a restricted pool, the four model families — frontier, balanced, fast, and specialist — still apply. If your administrator allows models from multiple families, automatic selection can escalate a complex legal analysis to a frontier model or route a quick rewrite to a fast one, all within the approved set. If only balanced models are permitted, the selector optimizes among those. The system adapts: fewer eligible models means fewer choices, but the scoring on capability, cost, and reliability still runs, so you get the best available answer from what's allowed.
Missions and agents work slightly differently. A mission runs on its own model pool, configured directly by the administrator, and can consult the same real-time reliability signals that chat uses. This means an admin might give a mission access to a different set of models than the ones available in your everyday chat. Between steps, a mission agent may even switch models — but again, only within its configured pool. The principle is the same: the administrator defines the boundary, and the system optimizes within it.
If automatic selection keeps choosing a model you didn't expect, check whether your plan or team settings limit the pool. A restricted set means the selector has fewer options. You can always pick a model manually — the choice holds for the rest of the conversation — or regenerate the last answer with a different model to compare.
Open a chat and send a message without choosing a model. Look at the badge under the answer to see which model was selected. Then open the model selector: the models listed there reflect your plan and team restrictions. Note which families (frontier, balanced, fast, specialist) are available to you — that's the pool automatic selection works with.
Your plan sets a baseline of available models; your administrator can narrow it further to approved providers. Automatic selection only routes within this allowed pool, ensuring data stays where your organization wants it. The four model families still apply inside the pool, so routing can escalate or simplify as needed. Missions have their own admin-configured pool. To see your current eligible models, open the model selector in any chat — that's exactly what automatic selection sees too.