Microsoft launches Critique & Council
Microsoft unveiled Critique — a multi‑model deep research system for M365 Copilot that scored 57.4 on the Draco benchmark and is positioned for production use. It also released 'Council', which runs multiple models on the same prompt to compare alignments and divergences for advanced workflows, and is available today. (x.com) (x.com)
Critique implements a two‑stage generator→reviewer pipeline that uses OpenAI’s GPT family to produce drafts and Anthropic’s Claude to perform rubriced reviews for factuality, sourcing, and completeness. (techcommunity.microsoft.com) Microsoft indicated Critique will operate as the default reviewer flow when Researcher’s model picker is set to “Auto,” and the release notes tie the feature into Copilot Cowork for persistent, multi‑step workloads. (constellationr.com) Council executes multiple models in parallel on the same prompt and returns side‑by‑side reports that surface agreements, divergences, and each model’s cited sources and rationale. (techcommunity.microsoft.com) Company benchmarking released with the announcement shows Critique outperforming every single‑model deep‑research system included in Microsoft’s DRACO comparisons, a result highlighted by testing write‑ups and news analyses. (testingcatalog.com) Rollout is scoped to Microsoft 365 Copilot’s Frontier program with tenant‑admin controls for turning Auto Critique and Council on or off and for restricting which external models tenants may call. (mc.merill.net) Microsoft positions Council as a procurement and compliance tool that lets organizations compare lower‑cost or domain‑specialized models per task while preserving provenance and audit trails for enterprise governance. (blockchain.news)
