I think they’re having some problems rolling it out on the copilot side. Are you able to send a copilot chat with the model?

I think they’re having some problems rolling it out on the copilot side. Are you able to send a copilot chat with the model?

Yes the copilot chat works just fine with both of the new claude 3.7 models. It's just not working in Roo.

i have the same issue, what is the solution? copilot chat works fine

I occured the same issue, awaiting for the solution. My copilot chat works fine.

same issue on roo, copilot chat can work well

Same here

me too

Same Issue, Cline and RooCode, Chat works fine. Was working yesterday morning, then there was the shutdown, but GitHub status says it resolved now.

Same here

The current system in use is the latest version of VSCode for macOS, it is uncertain whether there is such a problem on the Windows system.

I see there was a Upgrade that was Pushed a few Minutes ago on this Extension. i upgraded it, restart everything, but still encounter the same problem. - Windows 11

In my case It has been recovered for hours now, and it's working fine inside GitHub Copilot. The problem appears when using it with VSCode LM API.

In my case It has been recovered for hours now, and it's working fine inside GitHub Copilot. The problem appears when using it with VSCode LM API.

Your 100% right,

Only tested on Mac and Windows, Linux ssh server is also OK, you can try restarting or reinstalling the plugin

@EthraZa How did the performance of other models like 4o and gemini compare before and after the patch?

#1203 (comment)

Add support BYOK for Business and Enterprise

// Replace to support BYOK for Business and Enterprise
content = content.replace(/get isIndividual\(\){return this._info\.individual\?\?!1}/g, 'get isIndividual(){return !0;this._info.individual??!1}');

I just found out about the possibility of using Copilot LLMs with Roo theses days and it was already not working, so I don't know about speed changes.

To get it working I had to ask something to Claude 3.5 in the Copilot and click "Enable" when asked (again).
I was in the impression that Roo would use the selected LLM in Copilot and I tryed just 4o and Gemini. Today, after reinstall Roo and Copilot extensions I tryed Claude and now it is all working.

Tks

Microsoft changed the business model of copilot. I posted feedback to Microsoft https://github.com/orgs/community/discussions/156026
Please upvote that feedback if possible

https://github.com/DaydreamCoding/github-copilot-freedom

• Enable all models for LM API usage by removing authentication restrictions
• Enable BYOK (Bring Your Own Key) feature, which is normally restricted to Business and Enterprise users
• Enable CodeReview features for all users

The GitHub Copilot version 0.26.3 no longer supports the method of removing the "x-onbehalf-extension-id" to use Claude 3.7 Sonnet. I had to revert to version 0.26.2 to continue using it. Is it possible that GitHub Copilot will soon prohibit the use of this model via Cline and RooCode?

https://github.com/DaydreamCoding/github-copilot-freedom

@VietThuan I upgraded to 0.26.3 and it's working fine. Each time you upgrade the extension, you need to re-patch and reload the window.

@DaydreamCoding -- I just want to say that you are truly a remarkable human being doing a tremendous service to the community. I just updated to the recent versions of cline, github copilot and github copilot chat this morning and it was broken -- this time with a new error code -- and running your patch worked flawlessly. May the Divine's blessings be upon you and your family and loves ones. Much love to you and respect!

Verified working with:
Cline 3.11.1
GitHub Copilot 1.301.0
GitHub Copilot Chat 0.26.3

@gituser08jndmx Thank you, the best wishes to you.

Just wanted to give everyone an update. I’ve been a heavy user of Claude-3.7-Sonnet > VS Code LM API > Cline > VS Code for several months now. I’ve worked around rate limits (including juggling multiple GitHub Copilot Business/Enterprise subscriptions daily) and leaned on incredible community workarounds—shoutout to @DaydreamCoding.

But as of today, April 16, 2025, I’ve switched to Gemini-2.5-Pro-Preview-03-25 > Google Gemini > Cline > VS Code, and it absolutely blows away the competition.

The context window alone makes it superior to Claude in so many ways. I was previously forced to build React apps in ultra-small modular components just to stay within Claude’s limits. But with Gemini-2.5-Pro-Preview-03-25—despite Cline claiming it “does not support computer use” (it absolutely does, and it’s amazing)—I fed it 60 pages of documentation for Zoho CRM v7 and had it fully analyzed in under a minute. Within 5 minutes, I had a server-based custom portal built, giving my customers and contractors full access to Zoho’s platform.

This is a game changer like nothing I’ve seen before. Sure, cost is something to consider—but for me, replacing 8 Copilot Business subs at $160/month is a net win. More importantly, tasks that took 3 hours with Claude-3.7-Sonnet now take 30 minutes or less with Gemini.

The long context window changes everything.

Just wanted to share this for those of us pushing the edge of what AI-assisted coding can do. Nothing I’ve tried comes close to Gemini-2.5-Pro-Preview-03-25 > Google Gemini > Cline > VS Code.

Cheers, and God bless!

İs that working now?

İs that working now?

Yes, verified as of today, April 16, 2025. I guess I should've been less ambiguous in my comments above...

"But as of today, April 16, 2025, I’ve switched to Gemini-2.5-Pro-Preview-03-25 > Google Gemini > Cline > VS Code, and it absolutely blows away the competition."

"But with Gemini-2.5-Pro-Preview-03-25—despite Cline claiming it “does not support computer use” (it absolutely does, and it’s amazing)—I fed it 60 pages of documentation for Zoho CRM v7 and had it fully analyzed in under a minute."

"But as of today, April 16, 2025, I’ve switched to Gemini-2.5-Pro-Preview-03-25 > Google Gemini > Cline > VS Code, and it absolutely blows away the competition."

Awesome feedback, I’ve had a similar experience. The other thing it mention is the speed, it is way quicker than Claude.

I wondered if you have had a chance to try the new OpenAI 4.1 model, meant to be very good for coding and more affordable.

"But as of today, April 16, 2025, I’ve switched to Gemini-2.5-Pro-Preview-03-25 > Google Gemini > Cline > VS Code, and it absolutely blows away the competition."

Awesome feedback, I’ve had a similar experience. The other thing it mention is the speed, it is way quicker than Claude.

I wondered if you have had a chance to try the new OpenAI 4.1 model, meant to be very good for coding and more affordable.

No, I haven’t been using OpenAI’s 4.1 model. At the moment, I’m jumping back and forth between Claude and Gemini. Just fully vetting those two and learning their intricacies is a challenge in itself. I’ve never had much luck with OpenAI models inside my IDE, so I haven’t prioritized testing newer versions.

For me, it’s all about deeply understanding the tool — knowing how it fails, when it makes big mistakes, and where it falls short in terms of reasoning and debugging. Gemini is incredibly fast and has excellent coherence, but it’s not great at debugging.

I’ve been working on a very complex React app. Gemini and I spent 18 hours troubleshooting a dev dependency issue — specifically, trying to figure out why a local monorepo environment couldn’t be deployed to a server running the exact same environment without throwing dependency errors. Turns out, the issue was tied to a compression script that was packaging everything together. Somehow, the script was excluding files from the deployment package.

Twice, I performed a diff analysis with Gemini to compare the local package contents with the production ZIP — both times there were clear differences. I pointed this out to Gemini both times, yet it enthusiastically insisted that the differences couldn’t be the cause of the problem. After 18 hours of back-and-forth, I finally zipped the local deployment package myself, uploaded it to the server, and relaunched the app — and voilà, all dependency issues were resolved.

It was the first time I ever typed multiple berating insults and curse words to an AI in my IDE — furious that it had ignored repeated statements about the diff results. I even made it draft an email to Google API billing support, insisting that the charges for 18 hours of usage be refunded, with a full explanation that it was the AI’s fault. And get this — Google support actually accepted the email and is crediting my account. Wild ending to an infuriating journey caused by the simplest oversight.

So yeah — it’s all about learning how the tool behaves. Gemini is fantastic for certain tasks, and its coherence across a working codebase is superior to Claude. But Claude is marginally better at debugging — especially when used in ACT mode with Copilot. If you have the right Copilot subscription, the rate limits have improved significantly. I’ve gotten up to ~150 API calls before needing to switch accounts.

Anyway, hope that sums it up. Best of luck with your own experiments.