A team changes models but keeps the same number of requests. Costs increase sharply because prompts and answers are much longer. What should its monitoring include?

1. Input and output token usage by deployment and workload
2. Only DNS query volume
3. Only repository file count
4. Only the number of deployed resource groups