I Ditched Copilot Pro for Cursor—Here’s What I Learned After 30 Days

[TMRomain]

Select Topic Area

General

Copilot Feature Area

Copilot Agent Mode

Body

After a month on Cursor’s free student plan, I will be cancelling my GitHub Copilot Pro subscription after testing the new 'premium request' from github copilot. Cursor’s agent mode, pricing model, and day-to-day reliability fit my workflow far better, while Copilot Pro still feels opaque and rate-limited.

---

Where Copilot Pro Falls Short

Pain Point	Why It Hurts
Opaque rate limits	No live dashboard for remaining requests or which model you’re burning through.
Attachment mix-ups	I attach an HTML and a JS file, ask it to modify the HTML… and it edits the JS instead.Even if I try to remove the JS from the context and try it again it still edit the JS probably a bug.
Shallow agent mode	I can't get the agent mode to run longer when the task need it.
Rate-limited agent mode	Rate limit on agent mode even on GPT-4.1 base model, + the rate is very low and apply to all the Agent call so switching model don't work, I have to wait a few minutes to be able to use it again
Paywall on bigger models	Even on the Pro plan, I can’t use GPT-4.5, Claude Opus, etc.—why sell me “300 requests” if I can’t decide how to spend them? (Pretty similar to cursor Max but still)
Radio silence during outages	When Gemini was down pretty much all my work day today, neither the VS Code nor Copilot team said a word.

---

Why Cursor Won Me Over

Cursor Feature	Real Impact
Price-to-requests sweet spot	$5 more than Copilot Pro, but ~200 extra requests—effectively the same cost per call.
Deep agent mode	Watched Claude 4 build an entire API client, run tests, iterate for ~3 minutes straight—no manual nudges. And this whole process only took 0.75 request
Smart request accounting	An error or canceled response doesn’t burn a request. After 25 request to a model in agent mode (wich only burn one Cursor request) I can resume it and it picks up exactly where it stoped.
UI built for AI	In-editor errors pop bottom-left, leaving the chat pane uncluttered. Tiny tweak, huge focus boost.

---

Bottom Line

I love VS Code’s simplicity, and was all in with their previous pricing, and I’m excited that parts of Copilot are heading toward open source. But until the Copilot team:

1. Exposes real-time usage stats,
2. Lifts arbitrary model restrictions, and
3. Communicates during outages,

my $10/month is better spent on Cursor’s extra 350 Claude 4 or 250 Gemini 2.5 calls.

Hope this feedback helps GitHub steer Copilot back into the race, developers have options now, and we’re choosing with our wallets.

---

Have you tried Cursor or stuck with Copilot Pro? Let me know how it’s going for you!
Sorry for the title but I feel like this opinion need to get out there

[NhanPhanVN]

am not sure why this is the case, but for me github payed version is shitter than free cursor, every thing about it is just dumb, the error rate this through the roof, it cant even write working unit test for a complex file, cursor free on the other hand crush every task coming it's way , sql ? crush it with 1 try, unit test ? 1 hit , Js ? not even need to try . copilot with all its "microsoft" product is a joke.

[Akash1134]

Hi @TMRomain and @NhanPhanVN ,

Thank you for sharing the feedback. We have shared it with the product team.

Regarding VS Code real time usage monitoring, please check out this documentation: Viewing premium request usage

Regards,
Akash

[aris1009]

Is there an on-going discussion about raising premium request limits on pro plan? 50 requests a month seems like a bad joke at the moment. In comparison, Claude Code Pro allows ~45 requests per 5 hours at less than double the price.
I would get better mileage if I spent 10$ on raw API credits, without an active context window of 63k tokens on Gemini 2.5. My whole organization is moving away to either Cursor or Anthropic and I don't see Github acknowledging this issue anywhere.