GitHub in general is getting a lot of flak these days and I wanted to give my two cents.
Please comment your thoughts and if I got anything wrong or missed something.

GitHub Is Awesome

GitHub has always been awesome and it continues to be awesome. It is one of the wonders of the internet and I'm grateful every day that it exists. What it gives away for free and what it does for open-source in general should be appreciated.
Just like the rest of you I was very nervous when Microsoft bought GitHub. But I've forgotten about it ever since. The fact that massgrave is hosted on GitHub and Microsoft has not made a move on it, is a sign that they aren't just trying to throw their weight around (though who knows that it actually means). The vibes I get from MS these days is that they care about power users and developers. WSL, Windows Terminal, PowerToys, and VS Code, are all open-source, under very active development, and are simply awesome. Are there things to complain about? Yes, and you should complain about them. But understand that MS and GitHub are staffed by extremely talented engineers and managers that want to do good work and make things that people actually like. Sometimes difficult decisions have to be made. We all end up in such situations.

GitHub was not always part of Microsoft and was once a small startup. They have always had their own infrastructure that they had to maintain. GitHub has been saying for a while that they are working on moving GitHub to more reliable infrastructure on Azure. That can't be easy. Just because they now have the backing of Microsoft doesn't mean the system automatically has unlimited resources and shouldn't fail under the extreme load that GitHub has been under with the advent of competent coding agents pumping out code in ridiculous numbers (not that has to be the only reason they are having outages). Microsoft is not purposely ignoring GitHub or the engineers being dumb and ignoring obvious issues. They are working on it, I have no doubt.

Not that I don't think Microsoft is not at all to blame here. I would be surprised if that wasn't true at all. But I think it may be overblown quite a bit the reactions that I'm seeing.

I’m surprised how quickly some discussions jump from legitimate frustration to assuming incompetence or bad faith.

The transparency from the GHC team I think is really good. The livestreams, the posting on here and X/Twitter, the awesome open-source repos where everything is out in the open. It's fun to be part of the process and trust that the team is doing their best to make the product better. This is an extension of Microsoft and GitHub's appreciation for open-source and developers.

GitHub Copilot's New Pricing

GHC pricing until now was too good to be true and I appreciate that I was able to use a great coding agent (my preferred) for a long time for such a low cost. It is not hard to imagine that a user on the Pro plan using Opus 4.6 would be costing GitHub way more than $10 after just a few requests.

There will be a "preview bill" soon which will show you what your cost for April would've been with the new pricing system. But I don't think you should freak out if it's super high. It should be possible to significantly reduce the cost by being more deliberate about how you use GHC.

In general you should be more aware of tokens that are being fed to the model and offloading heavy token usage to cheaper models. For example, when doing code exploration, grabbing context from context tools such as Context7 and Tavily, it makes sense to make sure that goes in a subagent which uses a very cheap model. Also, I wonder if lowering the reasoning level (from my default which is high) to a lower reasoning level gives you significant token savings rather than just a speed bump.

Learn what the "prompt cache" is. Essentially it means that the tokens prior to the current message are priced at a much cheaper cost. See the middle column here. This cache can be broken though and is something that has to be kept in mind I think. For example, if you go back to an old chat and the cache is gone already (not sure how long cache is kept on the servers. Probably not long at all). I assume that if you switch models mid conversation that should break the cache and you'll be charged for every token at full price going into the new model. Note: Claude models charge an extra "cache write" price just for creating the cache. I'm probably getting some things wrong (or missing things) here about cache so let me know.

Why even have a subscription if you're paying full price for tokens? For Pro and Pro+, the reasons might be because it's just easier than setting up BYOK for each model (and even if you use OpenRouter there's an extra fee there) and you get the tuned prompts I'm assuming for each model in GHC. For an enterprise the reasons are more obvious what with the controls and analytics that GHC gives you out of the box very easily. Not to mention the GitHub Enterprise integrations. There are definitely reasons to stay on the subscription, while I don't see much reason to abandon the subscription. In the past BYOK seemed so expensive in comparison to premium requests so I never used it. But now I can explore more models and providers without feeling like it's more expensive. You can do both (as long as you use up your AI credits which you probably will).

Regarding the new high multipliers for the existing annual subscriptions, I don't think it makes much sense to keep the old plan. I'm curious as to why it's so much higher. I guess you can really abuse the premium request system? Especially for the kind of people that would pay for the annual plan? dunno. It doesn't seem to be simply double or some fixed number than the multipliers from before so there must be some specific logic behind it.

I like that I won't feel anxious anymore when sending small questions to GHC with expensive models (even though it will still be more expensive overall obviously).

When using the cloud coding agent (or code review) it seems that you will be charged for "actions minutes" as well as the model pricing. This is understandable because the agent runs in a GitHub Action runner. For public repos you have unlimited minutes. For private repos you get 2,000 free minutes, and an additional 1,000 if you pay the $4 a month for GitHub Pro.

There are no more free models which is a bummer because it was nice being able to use it for automations (with copilot -p or Copilot SDK) at no cost (though you still get unlimited completions and NES). Though the cheap models are super cheap so not a big deal. On the new models pricing table, GPT 5.4 Nano is mentioned. I'm excited to experiment with it and see what it can do (yes I can use BYOK but I'm lazy). Hopefully it actually comes to GHC soon.

Admittedly, I haven't used the other popular coding agent products much. But GHC always seemed the best to me. I like the compact UI, the VS Code integration, the policy controls, the JSON settings, and the fact that so much of the development happens in the open. Does it lag behind the absolute cutting edge sometimes? Maybe. But not enough that it has bothered me. Overall, the GHC and VS Code teams seem to be doing a lot right, and I’m excited for the new Agents app.

Keep it up GHC and VS Code team! ❤️