@osiewicz Sorry to ping you, but you've been doing some great Copilot work recently, and this would really improve the experience for Copilot users (for some of us it's the only officially sanctioned LLM provider at work, Microsoft's roots are clenched deep inside the corporate soul 😉). Zed's Agent is great, but the lack of Sonnet/Opus reasoning is a bummer. 👍

@reflectronic Hey! I've built it to test. Good job, thanks!

The only flaw i've found that sometimes it incorrectly reads stream, instead of one thinking block it pushes many.

What about for GPT models? With the same issue in Opencode it was discovered to be an issue with them as well. anomalyco/opencode#6864

What about for GPT models? With the same issue in Opencode it was discovered to be an issue with them as well. anomalyco/opencode#6864

yeah looking at the code this should solve it for all copilot models as the responses and chat completions endpoint were fixed

@reflectronic Hey! I've built it to test. Good job, thanks!

The only flaw i've found that sometimes it incorrectly reads stream, instead of one thinking block it pushes many.

can you say more? not worth merging if theres a bug

@reflectronic Hey! I've built it to test. Good job, thanks!
The only flaw i've found that sometimes it incorrectly reads stream, instead of one thinking block it pushes many.

can you say more? not worth merging if theres a bug

haven't encountered any other issues. works fine.

@reflectronic, @osiewicz Sorry to ping you, I know you're very busy, but would be really great to get this fix in. It seems rather straightforward and would have significant impact on the quality and usability of the Copilot provider integration. Currently the models are quite nerfed. 🙂

Hey, sorry to bump, but it's been a while, any news now that Zed and GitHub Copilot have a formal partnership? 😉

CC @yeskunall

Sorry for the delay in getting this over the line. I've updated the pull request to support the new 'adaptive thinking' in Opus 4.6 as well:

@reflectronic Thank you so much for moving this forward! 👍 The adaptive thinking switcher for the Anthropic models is a great addition as well. Would it be possible to add something similar for the OpenAI models too (reasoning effort, i.e. low, medium, high, xhigh)? 🙂

Hey @filipwiech if this is a feature request, it’s best to do so here.

Hey @filipwiech if this is a feature request, it’s best to do so here.

#51860 (comment)

@yeskunall , not sure how @guiopen 's question is a FR given the above interaction between @nesl247 and @somtooo:

What about for GPT models? With the same issue in Opencode it was discovered to be an issue with them as well. anomalyco/opencode#6864

yeah looking at the code this should solve it for all copilot models as the responses and chat completions endpoint were fixed

@yeskunall , not sure how @guiopen 's question is a FR given the above interaction between @nesl247 and @somtooo:

What about for GPT models? With the same issue in Opencode it was discovered to be an issue with them as well. anomalyco/opencode#6864

yeah looking at the code this should solve it for all copilot models as the responses and chat completions endpoint were fixed

In 0.228, the reasoning options do not appear for gpt models, just for Claude as the PR name suggest, so @somtooo comment could be a misconception

I created the FR from this suggestion:

Hey @filipwiech if this is a feature request, it’s best to do so here.

@udf2457 I said this should be a feature request because it is not something that was previously supported in Zed, and it clearly sounds like an improvement, not a bug. Do you have proof that supports otherwise? Then I will create the bug report myself!

+ thanks for chiming in as well @guiopen!

There is now bug/feature issue about the Copilot GPT model reasoning here #52140 for anyone wanting to follow it up.