Did GPT‑5’s Model Router Miss the Mark?

Did the model router for GPT-5 fail to live up to expectations?

Microsoft have announced a change for Copilot Chat where with GPT-5 as the new default model, we'll now be able to choose how much thinking we want Copilot to do. Either Auto, which is the default position where the model chooses, or now added, users can choose between a quick response or a deep-thinking answer, overriding whatever would have been the model's choice.

Similar options to this exist in other major AI chat services like ChatGPT or Claude, but up until recently Microsoft has steered clear of too many knobs and dials with Copilot. The big promise of the GPT-5 upgrade was that with the model router technology it would simply be able to intelligently select how much thought to give any request. But now with this addition potential complexity has been added where many users might feel driven to make a decision where in reality it could just be taken care of in the background.

From my perspective, I have rarely run into issues with just letting Microsoft 365 Copilot with GPT-5 enabled do its own thing, and wanting more granular control is not something I've heard from any clients.

I dream of a future state for Copilot where there are zero knobs or dials. Where it simply determines what to do based on context. No selecting agents. No choosing models. No calling out specific grounding content. Just me and a text box or voice interface that has complete insight and access to all work context and capabilities. An interface that harks back to the Google of the late 90s rather than a ChatGPT knock-off.

What do you think? Are you all for more control, or do you want the AI age to make things simpler all round?

See MC1192003 for more information. 🖼️ from the Microsoft message center post.

First posted on Linkedin on 12/05/2025 -> View Post Here