Which AI models does ELEV8 use?

Different tasks can have different needs. A model that is excellent for one class of reasoning may be slower, more expensive, unavailable in a required geography, or unnecessary for routine work.

Quality matters, but so do privacy, policy, reliability, latency, cost, tool support, modality, and the consequence of the task.

The customer or company may establish boundaries around approved providers, data classes, residency, confidentiality, or other requirements.

ELEV8's routing direction is to choose within those boundaries rather than treating provider access as permission to use customer information anywhere.

A new model can be evaluated against the relevant work before it becomes part of production routing. Testing should consider outcome quality, reliability, latency, cost, security and privacy boundaries, and other evidence appropriate to the capability.

Testing a new model does not mean immediately exposing customer information to it. Evaluation can use synthetic, Customer Zero, isolated, or otherwise governed paths before customer use where appropriate.

Because the model ecosystem changes quickly and the approved provider set can depend on customer policy, geography, capability, and current ELEV8 validation.

This page explains the operating principle. Where a customer needs exact provider commitments, those should come from the current applicable policy, configuration, or agreement rather than a stale marketing page.

Customer-provided provider accounts can be useful for bounded workloads or customer-specific requirements where supported.

ELEV8's commercial direction is not to make customers fund every routine workload through their own keys. BYOK is an additive path, not the default definition of the relationship.