AI coding is becoming increasingly more expensive. The companies that initially offered subsidies to attract customers are gradually phasing them out because they arenβt turning a profit. While users initially received a five-figure amount of tokens for $800 on Claude Code, that value has already dropped significantly.
In the long run, customers will have to get used to AI coding becoming more expensive. However, in the long run, there will certainly be a solution through better open-source models or through smaller, programming-focused models that run on less powerful hardware.
In any case, what is currently happening at GitHub is no coincidence and is also occurring among other industry giants that have previously subsidized their prices. No one can program with Opus/Sonnet for 10 hours a day for $200 a month anymore. That was nothing more than a dream, but tbh a really good one.
@toddsaunders A lot more of annoying questions instead of just doing his job. Its like the colleague your are coming to with an task and he comes every 5 minutes backs and ask the next question and at the end you are saying yourself "why i haven't done it on my own".
I believe that AI will enable us to make advances in research over the next twenty years that would have been unimaginable just a few years ago.
Curing chronic diseases and cancer, reversing the natural aging process, artificial organs, climate change, world hunger, and many other major problems. Itβs not that AI will solve all of this on its own, but AI is like the invention of the calculator for a mathematician. It boosts efficiency enormously and will lead to major breakthroughs being achieved in a short time, regardless of financial resources.
AI is leading us into an era where everyone has access to infinite knowledge, easily and quickly available. And that inevitably means that some people will be able to accomplish great things that would previously have been impossible due to resource constraint.
AI is the greatest enabler in history of humanity!
For Coding: GPT! For planning Opus and GPT together. Opus can detect gaps in the planning of GPT really well. But it tends to be a bit too strict when planning things on its own. GPT is the much more cost-effective option, as GPTβs tokeniser consumes around 40% fewer input tokens. This is particularly important for API-based coding tools such as Cursor, Windsurf, Lurus Code and a few others!
@elonmusk I would be delighted to see Grok eventually bring its AI software to Europe and have it hosted here by European partner companies. I know youβre not a fan of excessive bureaucracy in Europe but the rules here are what they are.
@sama A great model to create ultra realistic pictures! Hope to get a Sora comeback one day with the same quality for ultra realistic promotional clips
@DanielSmidstrup Opus 4.7 is great at planning and reviewing. GPT 5.5 is great for coding (much cheaper). You get the best results when working with both of them together, so they control each others work.
Iβm not sure weβll ever reach a point where developers can be replaced, because we still need creative minds who know how to build reliable, highly scalable software. However, Iβm fairly certain that if βtokens donβt matterβ, a single developer can achieve as much as at least 30 developers coding by hand.
With the right processes and an understanding of how AI works, you can develop extremely cool AI-powered software in record time. The cost of this might be roughly the same as what a developer in Europe earns (per user), but if one developer can do the work that would otherwise require many more, a company can still operate more efficiently and either save money or release more features.