Changes to Copilot have just rolled out and it is not good news...
Copilot premium request limits are now in effect.
Your premium request counter has been re-set to zero as of today, giving you a fresh start. Regular monthly resets will occur on the 1st of each month going forward.
Essentially a single busy day (even a few hours) will see you consume your allowance for the 'Premium' Modals. STRICT rate limits are also now in effect for ALL Modals.
Doubt they'll revert these changes. So just be aware of them.
This is going to have a significant impact if the prompt sent is large (which it is by default). This really is not looking good. @George any suggestions? This change by Copilot is essentially out to wreck the work you guys have been putting so much effort in to...
We are aware of the changes for GitHub copilot. They were in the works for some time now and well announced that there will be requests counting.
And the so called premium request have their multipliers depending on the model you choose.
So like the default GPT-4.1 model is still free and not counted as requests but others like Claude do count.
We still think that even per request limits are introduced in GitHub copilot they present a better deal than the per token counting but the user cases should prove that best.
Just wanted folk to know the changes if they begin to report failures. Not a great move by Copilot but who am I to state the obvious. Was nice while it lasted. Although I don't like Claude so much it is probably the best Modal available within Copilot (Gemini is up there too) and this is a big hit on that front. Chat GPT 4.1 is fine for basic tasks, and is still rate limited, but for anything slightly more complex it really is not so great. Those multipliers are vicious and obviously consume tokens incredibly fast if not used sparingly. May I ask how large the prompt is that Wappler is sending to them (ie the training data to make the LLM aware of Wapplers functionality), is this sent every time, or only initially, upon using the AI Assistant or is the response cached in any way to save tokens?
These changes were announced some time ago - so not really a surprise.
I started looking for alternates since announcement, and Windsurf and Cursor both look like a good alternative - much faster as well - but they are separate IDEs.
With these two, premium requests are cheaper - and the way their agent mode works, should usually consume few credits to get a lot of work done.
Not sure what Wappler's plan is to tackle this change.
We have just subscribed to Cursor ourselves so will have a play with it. Already hit limits of Copilot but thankfully have enough funds in Open Router. We won't be renewing our Copilot subscription.
Well there goes Cursor! They just played the Copilot card. Well Gemini is still going all guns blazing and with its introduction in to Wappler via the hard work and effort of the Team seems, for now at least, that for heavy complex intense workloads, Gemini is the way forwards. I'm no Google 'fanboi' but outside of Qwen and Deepseek's offerings Gemini is the most efficient and economical Modal available. I will add that Claude seems to have bounced back a bit this week, at least, is not as stupid as it has been in the past couple of weeks, was astonishing how quick Claude lost the plot (3.7 and 4.0) so they must have rolled back some changes..? For a while it really went off on a token binge returning pure and utter garbage. However it got better at apologising... Maybe that is a feature now?
Yes that is very unfortunate with GitHub copilot, hopefully as Google Gemini now rises with their great offerings copilot will take a lesson.
in Wappler 7.1 we have also added Google Gemini as a new AI provider, so you can now use Gemini 2.5 Pro or Gemini Flash 2.5 with their 1MB tokens large context window and reasoning capabilities.
And Gemini is actually giving pretty good results.
@George Really appreciate that George and wanted to highlight that above to those that did not know. Now Cursor has done the same as Copilot. Leaving Gemini a real contender for those of us that are 'Power Users'. The next month or two is going to be very interesting as the Chinese Modals start to come in to their own (more publicly). We have always been fans of these Modals. I'd say a good 80% of our workload is processed between Qwen and Deepseek, with Gemini Pro now exceeding our Claude usage, and ChatGPT well we just don't use them anymore after repeated terrible performance and output. Not sure how ChatGPT are going to respond but could have something special in the pipeline but who knows..? What a crazy time to be a Developer this really is!
I've started to use them (Claude, Gemini, GPT) together on really complex requests. I'll have one produce something, then give it to the next, and ask if they see any areas for improvement. Unsurprisingly, they each improve a little part. It's like having three employees that, when combined, are better than any one of them.
I have been experimenting with Windsurf lately. They had changed their pricing not too long ago. But I fear they will probably complicate things as well if they follow Cursor.
Copilot looks like a better deal compared to the messy Cursor pricing right now - but probably not.
I ended up using claude code for now with pro max subscription inside wappler/vscode paired with copilot and it's a powerhouse! Having the higher context window, multiple agents performing same task at the same time, oh and having the access to non-diluted models, is super efficient!
You simply /init your project, it knows your stack, I mean have fed it some propts from wappler's AI, and well, it performs tasks VERY well. It has steeper pricing point, but so do the providers like Cursor if you were to use their SONNET 4 MAX model. I have never hit the rate limit yet, and using GitHub - getAsterisk/claudia: A powerful GUI app and Toolkit for Claude Code - Create custom agents, manage interactive Claude Code sessions, run secure background agents, and more. i am able to swap between models and task with ease! The lower tier PRO MAX comes up £90, however, within the first few days of use, I based on Claudia's calculations, I have already used $113 USD worth of tokens if I were using API based pricing.
We are looking at the Pro Max subscription now. Especially as the access to Claude Opus seems pretty much unlimited (or at least very hard to hit the limits). Of all the Claude variants Opus is really really very good indeed. Thanks for the feedback @Evaldas
np. It also has a cool feature called orchestrator which takes huge context tasks, breaks them down and then only passes the needed context as sub tasks to the model you choose.
I tend to hit 80% rate limit with OPUS, very fast! Within first 40 minutes of session, I know because it switches to sonnet once 20% of tokens are left, however, even with 20% of tokens left, I didnt hit the limits that 5 hour window! Considering upper tear gives you 3 times the tokens it would be something I would eye on! But you can upgrade with difference at any point! But with good enough /init , sonnet 4 flies!
It created cluster of server connect modules for my needed SDK from a single well explained prompt! And just a few bits and bobs that I had to alter! super impressed
super impressed