Microsoft launches Critique and Council for M365 Copilot research

• Microsoft added two multi-model research modes to M365 Copilot: Critique splits report generation and review across different models, while Council runs the same prompt across multiple models side by side.
• According to Microsoft's Draco result, Critique scored 57.4 on the Draco benchmark, and the launch post says that is a 7-point gain over earlier Researcher versions.
• The reported Critique workflow is sequential: one model handles planning, retrieval, synthesis, and drafting, then a second model checks claims, structure, evidence, and citations before the report is finalized Critique details.
• Council appears aimed at comparison rather than review: the product demo shows several models answering the same research prompt in parallel inside the M365 Copilot interface.

Microsoft's new release is really two different multi-model patterns inside the same research workflow. In the Critique announcement, one model produces the research output, while a second model acts as reviewer. The reviewer is described as checking factual grounding, strengthening structure, and improving citation quality rather than generating from scratch. A separate product clip shows Critique surfaced directly in the Copilot interface as a research feature rather than a backend-only change.

Council takes the parallel path instead. The Council demo shows the same prompt executed across multiple models at once, with outputs displayed side by side in the Researcher experience. That makes it a comparison mode for prompt-level variance, whereas Critique is a staged handoff between generator and reviewer. Together, the launch shifts M365 Copilot from single-model answering toward orchestrated multi-model research flows.

The clearest performance claim is Microsoft's reported 57.4 score on Draco. In the benchmark post, that number is attached specifically to Critique, and the rollout summary says it is "+7.0" versus previous Researcher versions. The public detail here is limited, but the claimed gain is tied to the reviewer step: better factual accuracy, broader analysis, stronger presentation, and better citations.

The demo evidence also clarifies how Microsoft is differentiating the two modes in practice. The video teaser for Critique ends on "Draco Benchmark: 57.4," reinforcing it as the quality-focused path, while Council interface shows Council splitting one Q2-summary prompt into multiple concurrent outputs. For engineers, the implementation signal is less about a new base model and more about orchestration: M365 Copilot is exposing multi-model routing patterns as product features.