Skip to content
Back to feed

November 20, 2026 · AzureIn preview

Public Preview: Multi-turn evaluation in Microsoft Foundry

Multi-turn evaluation is now in public preview in Microsoft Foundry. You can assess complete agent conversations against your quality standards, not only individual responses.

A response can look correct on its own and still be part of a conversation that fails the user. For example, an agent can repeat the same failing action across several turns and leave the request unresolved. Turn-level scores do not show this pattern.

With multi-turn evaluation, you set the evaluation level of a run to score each turn or the conversation as a whole. Conversation-level evaluators, such as task completion, customer satisfaction, coherence, groundedness and tool use quality, find problems like incomplete task resolution, user frustration and tool-call regressions. You provide conversation data as JSONL with a full messages array, including tool calls and tool definitions, and run the evaluation with the Foundry SDK.

Learn more.

Get changes like this in your inbox

Free, every Monday: what changed across AWS, Google Cloud, Azure, Cloudflare and the AI providers. We email you a link to confirm.

Free. Unsubscribe any time.

All Microsoft Foundry changes