A company evaluates a new foundation model for a customer-facing assistant. Which two criteria should drive selection beyond a public benchmark score? Select 2 answers.

1. Only the lowest input token price regardless of answer quality.
2. Measured latency, cost, context requirements, and supported deployment geography.
3. Quality and safety on representative domain prompts and important user cohorts.
4. Only the largest advertised parameter count.
5. Only the provider logo shown in the console.