Why does GPT-4.1 under Copilot’s agent mode seem so dumb?

[facelezzzz]

Overall, I think it’s quite useless. Most of the time, it feels like it’s running in editor mode rather than agent mode.

It is far behind Claude Sonnet.

• VS Code Version: Version: 1.101.1
• OS Version: Mac 15.2 (24C101)

Steps to Reproduce:

[vs-code-engineering]

Thanks for creating this issue! We figured it's missing some basic information or in some other way doesn't follow our issue reporting guidelines. Please take the time to review these and update the issue.

For Copilot Issues, be sure to visit our Copilot-specific guidelines page for details on the necessary information.

Happy Coding!

[wallyfoo]

I think it's a clever ploy to generate revenue from premium requests. My experience is the same as yours. Chat GPT 4.1 introduces more errors most of the time than it helps me to resolve. If I switch to Claude Sonnet 4, it's usually one and done, but absolutely plows through my premium request budget.
VSCode 1.101.2
MacOS 15.5
Copilot Pro

[facelezzzz]

It has been very disappointing for me and many others. I think a lot of us will cancel our subscriptions as soon as possible.

Also, I still don’t understand why o3, which is not even considered the standard model, is priced the same as GPT-4.1, even though it seems more powerful.

[weareoutman]

It seems GPT-4.1 can't understand some tool input schemas (especially anyOf), it always reports "failed due to a schema issue" or "not passing validation" when trying to use some of my MCP tools. However, it works when switch to Clause Sonnet 3.5/3.7/4.

[digitarald]

@weareoutman can you open issue for the specific servers, please?

Insiders has some 4.1 prompt and editing improvements; can you check against those. We are working on improvements to 4.1 pausing less for clarification, but if that happens a "Continue" usually keeps it flowing.

[weareoutman]

@digitarald That MCP server is private and deployed locally. I'll try insiders later.

[facelezzzz]

UPDATE:
I unsubscribed Copilot Pro Plan. I mainly use Gemini Cli now which is better than Copilot.They offer Gemini Pro 2.5 for free, while I never use GPT-4.1 mode because it is useless in agent mode.

Copilot Pro plan give me 300 Gemini 2.5 Pro request per month request but charge 10$,Gemini Cli give me 100 Gemini 2.5 Pro request per day for free.

And Gemini Code Assistant Pro Plan can give us 1000 Gemini 2.5 Pro request per day and it charge you only 19$ per month

I think the developers of Copilot are already doing a great job, but they’re held back by the company’s business strategy and the underwhelming GPT-4.1 model.