I share the same sentiment. I know many people praise Kagi, and I respect the effort behind it too. I tried it for three months and realized it was not for me. Google works just fine for my needs.
The Comfort Crisis: Embrace Discomfort To Reclaim Your Wild, Happy, Healthy Self by Michael Easter.
Never Finished: Unshackle Your Mind and Win the War Within by David Goggins.
These two books stuck with me. We are often too comfortable with being comfortable. There is nothing wrong with that, but real growth happens when we step outside our comfort zone. We are far more capable than we think.
You're comparing a language with a framework. Better comparison would be Rust (with your choices of libraries) vs Ruby (with your choices of libraries).
If you want to compare with Rails, you need to compare with battery included Rust frameworks with equivalent batteries and convention.
I generally prefer Sonnet as comparison too. Opus, as good as it is, is just too expensive. The "best" model is the one I can use, not the one I can't afford.
These days, by default I just use Sonnet/Haiku. In most cases it's more than good enough for me. It's plenty with $20 plan.
With MiniMax, or GLM-4.7, some people like me are just looking for Sonnet level capability at much cheaper price.
Are you using GLM-4.7? I've just spent a fortune on Opus, and I heard GLM was close -- but after integrating it into cursor, it seems to spin forever, loose tool use, and generates partial? plans. I did look into using it with the claude cli tool, so it could be cursor specific -- but I havent had the best experience despite going for the pro plan with them. Any advise on how you're using GLM effectively? If at all
At the moment Opus is the only model i can trust even when it generates "refactoring work", it can do the refactoring.
I’m on the Lite plan. For coding, I still prefer Claude because the models are simply better. I mainly use CLI tools like Claude Code and OpenCode.
I’m also managing a few projects and teams. One way I’m getting value from my GLM subscription is by building a daily GitHub PR summary bot using a GitHub Action. It’s good enough for me to keep up with the team and to monitor higher-risk PRs.
Right now I’m using GLM more as an agent/API rather than as a coding tool. Claude works best for agentic coding for me.
I’m on Claude $20 plan and I usually start with Haiku, then I switch to Sonnet or Opus for harder or longer tasks.
> I did look into using it with the claude cli tool, so it could be cursor specific
Claude Code with GLM seems ok to me, I just it use it as a backup LLM if in case I hit usage limits but for some light refactoring it did the job well.
Are you also facing issues with Claude Code and GLM?
No matter the price they're far cheaper than a developer and opus / gemini 3 pro are both at a level where they're really useful pair programmers and opus at times can be given a spec to implement and it will do it after 30 minutes with no input from me.
Per successful tasks. The result are mixed. Like you mentioned, it can be cheaper but only in some use cases. I'm only on the $20 plan. If I use Opus and it's not as efficient for my current tasks, I'll burn through my limit pretty fast. Ended up can't use any anymore for the next few hours.
Whereas with Sonnet/Haiku, I'm much more guaranteed to have 100% AI assistance throughout my coding session. This matters more to me right now. Just a tradeoff I'm willing to make.
> Claude Opus 4.5, our frontier coding model, is now available in Claude Code for Pro users. Pro users can select Opus 4.5 using the /model command in their terminal.
Opus 4.5 will consume rate limits faster than Sonnet 4.5. We recommend using Opus for your most complex tasks and using Sonnet for simpler tasks.
Use Claude Opus in Antigravity. Google is very generous with the limits. The best part is, if you hit your limit, you can switch to Gemini Pro High.
I think Google is able to do this because they host Claude on their own TPUs in their datacentres (probably for Vertex AI customers). So they can undercut just about anyone include Anthropic on costs!
No matter which model you start with, having the other frontier model as a backup is fantastic. Essentially you’re getting 2x the limit.
Part of it is because Ruby imo, have a very nice syntax. With type annotation, it's becoming "ugly", a lot more verbose. It's no longer English-like. I do agree type have some advantages, but we need to get the DX right.
I've been using Ruby for more than 10 years now, and I only started using LSP recently. To me it's a nice addition but I can live without it. Type is just one of the tools, not the only one imo. Not trying to sound negative but type is becoming more like a hammer analogy nowadays.
And it's not limited to Ruby. Javascript, Python, all similar languages. Not everyone is a fan of type. We won't reach consensus imo and that's ok.
> With type annotation, it's becoming "ugly", a lot more verbose. It's no longer English-like.
In our codebase that uses Sorbet I find this is really only true at function boundaries. Within a function it is pretty rare that anything needs to be spelled out with inline annotations to satisfy the compiler.
This is my biggest irk about Sorbet: because its signatures are wordy and because it can't infer the generic type of a private method, it slightly pushes you towards NOT extracting helper methods if they are going to be 2-5 lines. With Sorbet annotation, it'd easily become 10 lines. So it pushes towards bigger methods, and those are not always readable.
If only private methods would be allowed not having typing at all (with a promise of not being used in subclasses, for example), and Sorbet would be used mostly on the public surface of classes, it'd be much more tolerable for me.
1. Put bunch of bank statements pdf in a folder, give a deterministic output for each pdf. Then ask Claude Code to do whatever I want. Good enough.
2. My preferred approach is similar to above but ask it to write a script instead, eg in Ruby. That way I have proper test, 100% guarantee it'll work and no regression. AI is non deterministic by default so asking any kind of agent to give a deterministic output seems unreliable to me. In the end I've turned it into a CLI, and been using it till now.
That's how I use AI. Indirectly to get what I want. Chat, CLI, it's all just a medium.
You still need ridiculously high spec hardware, and at Apple’s prices, that isn’t cheap. Even if you can afford it (most won't), the local models you can run are still limited and they still underperform. It’s much cheaper to pay for a cloud solution and get significantly better result. In my opinion, the article is right. We need a better way to run LLMs locally.
You still need ridiculously high spec hardware, and at Apple’s prices, that isn’t cheap.
You can easily run models like Mistral and Stable Diffusion in Ollama and Draw Things, and you can run newer models like Devstral (the MLX version) and Z Image Turbo with a little effort using LM Studio and Comfyui. It isn't as fast as using a good nVidia GPU or a cloud GPU but it's certainly good enough to play around with and learn more about it. I've written a bunch of apps that give me a browser UI talking to an API that's provided by an app running a model locally and it works perfectly well. I did that on an 8GB M1 for 18 months and then upgraded to a 24GB M4 Pro recently. I still have the M1 on my network for doing AI things in the background.
I’m curious to hear more about how you get useful performance out of your local setup. How would you characterize the difference in “intelligence” of local models on your hardware vs. something like chatgpt? I imagine speed is also a factor. Curious to hear about your experiences in as much detail as you’re willing to share!
Local models won't generally have as much context window, and the quantization process does make them "dumber" for lack of a better word.
If you try to get them to compose text, you'll end up seeing a lot less variety than you would with a chatgpt for instance. That said, ask them to analyze a csv file that you don't want to give to chatgpt, or ask them to write code and they're generally competent at it. the high end codex-gpt-5.2 type models are smarter, may find better solutions, may track down bugs more quickly -- but the local models are getting better all the time.
This subthread is about the Macbook Air, which tops out at 32 GB, and can't be upgraded further.
While browsing the Apple website, it looks like the cheapest Macbook with 64 GB of RAM is the Macbook Pro M4 Max with 40-core GPU, which starts at $3,899, a.k.a. more than five times more expensive than the price quoted above.
I was pleasantly surprised at the speed and power of my second hand M1 Pro 32GB running Asahi & Qwen3:32B. It does all I need, and I dont mind the reading pace output, although I'd be tempted by M2 Ultra if the secondhand market hadn't also exploded with the recent RAM market manipulations.
Anyway, I'm on a mission to have no subscriptions in the New Year. Plus it feels wrong to be contributing towards my own irrelevance (GAI).
You can also use z.ai with Claude Code. My workflow:
1. Use Claude Code by default.
2. Use z.ai when I hit the limit
Another advantage of z.ai is that you can also use the API, not just CLI. All in the same subscription. Pretty useful. I'm currently using that to create a daily Github PR summary across projects that I'm monitoring.
> mostly web developers who care about this in the first place.
I’m not sure what you mean by this. We care about our users and how they use our websites. JavaScript is everywhere and has been the de facto frontend standard for the past few years. Supporting no-JS is starting to feel like supporting a new browser. As much as I’d like to, from a business and product point of view, the numbers are just too small for us to even consider it.
I didn't imply that all web developers care about it, but that most of the people who care about it are web developers. I won't deny that it's still a minority.
Which LSP are you using? I'm using both solargraph and ruby-lsp and both works fine by me (in neovim).
Although those who really care about LSP support usually will use RubyMine IDE instead. Some of my colleagues are going that route, and they're mostly coming from Java (or similar background)
I'm not really "using it", I'm just trying every now and then, and I keep encountering errors, hangups, and lack of functionality. Now I've tried ruby-lsp, and it just sits there on "Starting Ruby LSP...\n"
Couldn't even install Solargraph, once it errors out with 'Kernel#require': cannot load such file -- yard, other time it installs, but "solargraph scan" fails in runtime with "missing gem date" error.
Sorbet doesn't even work in VSCode, some bugs are over 5 years old.
But yeah, downvote my original post, because apparently all of the above is obviously my fault.
RubyMine was paid until recently, now it's free only for non-commercial use. It's also not really suitable for small scripting.
Historically, one insanely huge advantage of Ruby was that it was pre-installed on macOS'es, but I think they've stopped doing that since some macOS version.
reply