POV: you're still using GitHub Copilot after June 1st, 2026

We've been reduced to mini and haiku, might as well go back to stack overflow and get others to answer my questions with their LLM queries...

Or switch providers?

Sadly not my call

The spend is actually insane. The simplest of prompts with the agent set to auto is $5. Its insane

Indeed. I get that the API pricing is expensive but much of it in GitHub Copilot is due to bad context management like stuffing all skills and MCPs in there on every request

Yeah, I agree it has to be something like that. Just running Claude code on my personal computer the cost is a fraction. I can run the $100 plan without issue for close to the full month. At work its ~$50 / day

@dedene To be fair, Claude's system prompt is permanently cached on Anthropic's server whilst GitHub's prompt is re-cached after TTL. Meaning, that alone can cost you $30+ bucks a day. I think AI's will get crazily expensive the next years, so invest in local AI whilst you still can...

@quasi_mortal @dedene Cost is the underrated local-vs-cloud driver. Local does not need to win every reasoning benchmark to be useful. It just has to handle the repeatable/context-heavy work cheaply, while cloud handles the hard frontier calls. The split is the product.

@PilsZehn @quasi_mortal This 🎯

im on the student plan, a single request exhausted my entire monthly quota

that’s rough

Time for a good ol cancel of copilot.

Yup, time to move on..

@temprlflux The only place people can move on are Chinese models. Because all the big 3 (Anthropic, OpenAI, Github Copilot) in enterprise world switched to API pricing.

@dedene It was inevitable. I think it will be a mix of open-source local models and sota models running within harnesses. Grok Build lets you use any model, so you can use SOTA for hard tasks and delegate to local for simple things.

Yeah I felt the same when Opus 4.5 destroyed my $20 in a single prompt... using the api key... in 2mins.. and did not even finish the task.. I gave. I decided I'm never going back to Anthropic again... without a subscription on api costs.... I learnt the lesson quickly 😅

Yup, we've all been there 😅

could be an option

👌🎯

Yeah, and Mythos is going to be a nuke...

It's time to switch to DeepSeek

🎯 or Composer-2.5 or Grok Build

@Cybelak i tried grok build... well... its not a good model for coding imo.

@Cybelak

@Cybelak well, I not gonna be the one losing my time giving training data so they can fix it. not that poor :D

@Cybelak anyway, my silly company just went copilot with shared pools for the whole company... i can even know how many credits i spend... this is going to be fun... ahahhahaha

@Cybelak that’s gonna be a fun experiment 😅

it's REALLY is like this. Within only 4 HOURS of casual (!) coding I used up my entire MONTH of Copilot Plus!!! That's insanity @github



"Ok now commit our changes and push to github" = $3.5

Bro i mess it up 😩 100 usd in a single day

@DomMuker

I exhausted my whole AI Credits allowance for the pro plan in less than 2 days, which previously took me about minimum of 15 days with my full power working, WTH did you do ? @github @GitHubCopilot

They’re also punishing people who paid annually by increasing how much usage prompts take up. I switched to monthly but since I’m still technically an annual subscriber I’m being punished.

Yes, that's really shitty. It's one of the reasons why when it comes to AI subs I stopped to subscribe yearly, even if it will cost me more.

Before june, use to do 10-20 copilot code review per day. After june, 10-20 copilot code reviews per month

yeah the token consumption went through the roof... we've switched to Deepseek V4 Pro and are way more satisfied 😎

That is sooo good! AI is not capable of that level of brilliance! 😂

🫡

Exhausted my quota in 2 days of light work 😂😂 Time to try them Chinese models + my brain 😂 It was fun 😂

@Drachs1978 GitHub copilot was built entirely for a different era of AI spend.

Yea, GH needs to fix the tool loading bug ASAP. Copilot loads all the tools on all the requests, even when they are not needed, eating 40k average (for me) per message (not enter press, agent message in a thread)

Indeed, now they're fully usage based they really need to work on context management and efficiency

I am thankful that gh copilot has been subsidized for so long. have saved thousands of $ over the last two years maxing SOTA models each month. It was good while it lasted.

I noticed. Cursor free is actually better.😂

And it doesn't even do proper build able code in a serious task. I told him to refactor and fix errors in 4 blazor files and he spend 40% of the month's free usage in that and didn't fix anything. 😒 Pathetic. They are going straight from anyone can use a.i tools to only millionaires will now can. Well screw it. Hiring a developer and paying salary will be cheaper!😅

And here I am spending less than a dollar a day on DeepSeek v4 Flash

Vibe coding is dead, company can restart hiring Devs.

I cannot use the 1500 requests by the whole month of May, but got exhausted on the first day of June.

Claude Mythos

Lol nice banger sir

Thank you sir 🫡😅

Not to sound elitist, but with the amount of good options out there, I'm blown away people still use copilot.

omg what a banger 😂 over 1m views

so crazy 😂 believe it or not, first time I actually used Canva to edit the text on that classic war spending clip to create something original

I cancelled mine last month but still had 6 days of June and the entire June "credits" allocation given to finish up. It finishes after 3 hours with sonnet 😂 Last month I was rage wasting my May allocation every day and I still had 40% by the end of the month.

Copilot is actual dogshit now. The stuff I did in a day and I ran out of my monthly credits. It's nearly as bad as the free plan now. In the past, I could go a month doing whatever I wanted using Claude Sonnet.

🤣

Copilot CLI is very good

That is absolutely true! Very underrated cli 🔥

Are thise chinese AI models really better? They are cheaper than claude. Never used them though

I think the DeepSWE bench tells the whole story

Interesting. So its good that I didn't use them😂

@dedene There are different benchmarks. For instance, on LMArena they're really good.

@dedene Oh is it? Have you tried any?

@dedene Currently using the open source models via Opencode.

@dedene Oh okay