Copilot Cowork model choices need a task-level cost test
Reviewed by Helen Jones
GPT-6 Astra is now available as a model choice in Microsoft 365 Copilot Cowork. In a 23 September hands-on article, Vesa Nopanen used it on a real writing task with supplied research and a reusable writing skill, while explicitly avoiding any claim that Astra had already beaten Auto or another model.
A new model badge makes it tempting to assume the strongest-sounding option will produce the best finished work. Microsoft’s own guidance says model and reasoning-effort choices trade speed, depth and Copilot Credits, so a more expensive run still has to reduce correction and review work. For Copilot adopters, the useful measure is not which first answer looks most impressive. It is whether the same task reaches an acceptable result with fewer missed requirements, factual errors, retries and minutes of human editing.
Analysis
Run one real task twice with the same source material and acceptance criteria: once on Auto and once on Astra. Record elapsed time, credits, correction rounds and whether you accepted, revised or restarted each result.
Source note
Pulse published by Collab365 Spaces, reviewed by Helen Jones on . Cite as "Copilot Cowork model choices need a task-level cost test", Collab365 Spaces.