A three-person team wants to call GPT-4o from Foundry to power a low-traffic internal helper bot. They do not want to provision or manage any compute instances, and they want to pay only for the tokens they actually consume. Which deployment option in Azure AI Foundry best fits this need?
Select an answer to reveal the explanation.
Short Explanation
Think about what this team actually needs: nobody on a three-person team wants to babysit servers, and traffic is light enough that a fixed always-on deployment would mostly sit idle burning money. The pay-as-you-go serverless option in Foundry is built for exactly this - you call the model, you get billed for the tokens you used, and there is no VM or cluster sitting behind it for anyone to patch or resize. A provisioned endpoint sized for peak load solves a different problem: it is for teams that need guaranteed capacity and are willing to pay for it around the clock, which does not fit a low-traffic internal tool. Standing up your own Kubernetes cluster is even further from 'no infrastructure to manage' - now you own the nodes too. And a queue that only processes once a day is fine for overnight reports, but it would leave a chat user staring at a blank screen for hours. When cost should track usage and the team wants zero servers to think about, pay-per-token serverless is the natural fit.
Full Explanation
The correct answer is B. A serverless API deployment in Foundry lets the team call the model over a managed endpoint without provisioning or maintaining any compute, billing only for the tokens the low-traffic bot actually consumes, which matches both stated constraints. Option A is incorrect because a managed compute endpoint requires selecting and paying for a fixed instance size that keeps running (and costing money) whether or not requests arrive, which is the opposite of pay-per-token. Option C is incorrect because standing up and autoscaling an AKS cluster means the team must patch, monitor, and size nodes themselves, directly contradicting the requirement to avoid managing compute. Option D is incorrect because a once-daily batch endpoint introduces hours of latency between when a request is submitted and when a result is available, which does not suit an interactive helper bot that needs a near-immediate reply. For a small team with unpredictable, low-volume traffic, the serverless pay-as-you-go model removes infrastructure ownership entirely while keeping cost proportional to actual usage, which is exactly the tradeoff this scenario calls for.