Microsoft upgrades Copilot agents to GPT-5.2 model

Microsoft has updated its declarative agents in 365 Copilot to run on the GPT-5.2 model. This upgrade improves reasoning, planning, and the structure of outputs in complex work scenarios.The change rolls out gradually across tenants worldwide and finishes by the end of March 2026. Users relying on agents for prompts with tool use or multi-step logic may notice subtle shifts in behaviour.

Copilot agents previously faltered on tasks needing deep planning or reliable formatting, often requiring users to rewrite prompts or intervene manually. That unreliability kept automation shallow, limited to simple queries in email or Teams.GPT-5.2 pushes agents toward handling real workflows end-to-end, like coordinating tasks across apps without constant fixes. The result is automation that sticks, cutting the time spent firefighting scattered processes.