Github Copilot is a scam

[binidom]

Select Topic Area

Question

Copilot Feature Area

General

Body

How to understand this, I began to notice that sonnet 4 is becoming very stupid and cannot solve simple problems. I decided to ask him what version it is and I was very surprised and discouraged that this is possible in a large company. I have version 4 in the settings, and for this money they are slipping me 3.5. Guys, have you accidentally become impudent there, doing this?

[binidom]

This is so mean and no one talks about it. I just struggled for 2 hours and couldn't figure out why this wouldn't be solved, gemeni solved it in 20 seconds. And I'm sitting here and I don't understand, I have sonnet 4 enabled and it can't do anything. I started to think maybe it was just me and then I asked him and got an answer. That I was using crap that was outdated and made in such a way that you wouldn't know until you asked. Who is he? How can Microsoft do such a mess?

[binidom]

[binidom]

I am creating a commercial product, I specifically paid 40 dollars to get sonnet 4 and get 3.5. I don't care about this model, it is not suitable for programming, it is a scam, missed deadlines, I lost money because of this situation, how can I get justice, who has the same problem?

[karadHub]

[binidom]

Sorry, brother, but you are deeply mistaken in your stories. If this is not there, then what is on the screen?

[karadHub]

May be my version is older.

[binidom]

Of course it's outdated, after it there was also version 3.7

[binidom]

And I didn't come because it says "Sonnet 4". It really is! What are you talking about, OpenAI? Did you even use anything for the code? Nothing can write code like "Sonnet 4". And it's on Github Copilot. But he silently switches to 3.5 and you can really feel it, he turns into a complete idiot

[QMS85]

If you switch between different models, you will get different outputs
& if you really don't want to use Sonnet 3.5 well then you can simply disable it, under your co-pilot settings.
You can also improve & refine your prompts, by giving co-pilot personal instructions, click on the menu option (the ellipsis)
Alternatively you can check this out:
https://learn.microsoft.com/en-us/training/paths/copilot/

[Wcpilot40]

That's the whole problem, I don't switch between models, why? I bought Soonet 4, the rest is not needed, because it's nonsense. Only Gemini is on the level, but I buy it from another provider. The problem is that in the settings everything is disabled except Sonnet 4, but it silently switches to it. I noticed this by accident, we are being deceived!

[atharhive]

I recently saw a research that says: vibe coding made expert programmer 20% slower!! i think you should check it
here : https://arxiv.org/abs/2507.09089

[binidom]

Today it works correctly and shows sonnet 4. Before that there were 3 days of complete disappointment, it couldn't do anything, I just couldn't understand what was going on. It's very difficult to predict business when such leaps in artificial intelligence occur. I had to refuse several clients because there is no alternative at an adequate price and quality. I already started looking at kimi2 today, but it thinks for a long time and is not what they say about it.

[binidom]

Again it became 3.5 and again it is stupid, how long will this continue, I pay for 4, in my settings everything is turned off except for version 4. What the hell, I have version 3.5? How much longer will this chaos continue?

[Sharma-Mayank15]

I’m building commercial software and rely on Copilot’s coding intelligence daily. I specifically subscribed for Sonnet 4 — yet I repeatedly experience silent downgrades to 3.5, even though my settings have every other model disabled. The drop in quality is obvious: hallucinated code, incomplete suggestions, and outdated syntax.

This isn’t about prompt engineering or expectations. It’s about getting the product I’m paying for. Copilot doesn’t notify users when it falls back to a weaker model. That’s misleading at best, and dishonest at worst.

If there are capacity issues or model switches, say it clearly. Don’t hide it. As a paying user, I deserve transparency, especially when my work depends on it.

Until this is fixed, I’m looking into alternatives like Gemini and Claude, which are at least consistent and upfront about the models they use.

Please fix this — model switching without notice is unacceptable in a professional tool.

[WCPILOT7]

The support says that you don't have a Business subscription and therefore support doesn't work for you and we can't give you an answer why this is happening

[ijash]

Hey microsoft! I have paid for a year, Please reset my premium requests to zero. I feel financially exploited. This is already close to 100% and it's crazy. Plus, there are so many errors, making us undo frequently.

[Omninovas]

Why you MUST NOT PAY FOR THIS SUBSCRIPTION? What they do? They make this AI really cool when it is free. Once your trial version is expired and you pay for it, it suddenly becomes stupid so your code can have a lot of mistakes. And the more you will want to fix the code the more you will need to pay for it. So you will end up spending much more money when you’re having the subscription because the AI is programmed to be stupid when you start paying to never fix something completely. Therefore, you can send more request and they are making more money and nothing is going to work.

[KnightNiwrem]

This is a known issue with the Sonnet 4 model itself, and is unlikely to have anything to do with Github Copilot. See: https://www.reddit.com/r/ClaudeAI/comments/1m0baxw/claude_sonnet_4_claudesonnet35/

Even calling Sonnet 4 against the direct provider produces such behavior. There is no financial incentive for Github to not give you what you ask for, since the Anthropic provider charges exactly the same pricing for Sonnet 3.5, 3.7, and 4. See: https://docs.anthropic.com/en/docs/about-claude/pricing

[WCPILOT7]

So, it's not that he writes 3.5, but that he becomes very stupid, you can really feel it.

[KnightNiwrem]

So, it's not that he writes 3.5, but that he becomes very stupid, you can really feel it.

That, too, is a reported issue of the provider, itself. See: https://www.reddit.com/r/Anthropic/s/cxhvUtpuY2

And many more similar reports with a quick search. It is expected that everyone will experience the same, if using the model directly from the provider also produces that result.

[binidom]

This is complete nonsense, they even immediately withdraw 20% when replenishing by 39 dollars. I replenished today, and my report became 20%. What does the cloud have to do with it? Github is engaged in deception. I can't finish the project, I feel like I use GPT 3

[binidom]

I activated Sonnet 4, everything is disabled in the settings except Sonnet 4. I don't care what's going on there. If you can't guarantee what you're charging us for, then don't offer us this service! THIS IS A REAL SCAM!