Hacker News Reader

Comment by o_m | Hacker News Reader

I used to main Claude, but I can't stand how it writes. I feel like I'm wasting too much time trying to decipher the output. Adding writing rules does not seem to work. Now I use GPT 5.6 Terra high fast mode, with Luna for everything else. I might consider using Sol for planning. I can't stand using Sol or smarter models for coding, because they will eventually try to rewrite everything in the codebase.

I also don't want to use the Claude Code and Codex agent harnesses. The good thing with Codex subscription is that it can be used in other harnesses, unlike Claude. As far as I know, only Anthropic has this restriction.

View on HN

I recently switched over to GPT from Claude once my subscription lapsed, it's too early to tell, but I do appreciate how more straight to the point Sol 5.6 seems to be, vs. the insane wordslop machine Opus tends to be.

It's really strange because when Opus 5 released, there were some that pointed this out, but a bunch simply said it was the best and as good as Fable etc etc.

but for me, it caused me to get very demotivated and avoid interacting with the model, at least when using Claude Code.

Funny enough, about 8 months ago I switched to Claude mainly because of MCP.

Now I’ve gone back to Codex because I simply find the ChatGPT models much more comfortable to work with. I spend less time fighting the model, correcting its direction, or re-explaining what I meant.

Me too. Claude Opus 5's English is insufferable. Opus 4.6-4.8 was more reasonable.

I've moved to Codex 5.6-Sol. Much saner English, much better at execution, and gets stuff done in a matter-of-factly kind of way (Claude Code is a mess these days -- it gets things wrong and goes around in circles).

But I'm harness agnostic and am not locked in. I just keep my issues in Kata Tracker (https://www.katatracker.com/) and switch harnesses/model when I need to.

Being loyal to a particular model/harness seem unwise to me.

> Me too. Claude Opus 5's English is insufferable. Opus 4.6-4.8 was more reasonable.

I suspect this degradation is happening because the AI labs are using the LLM's output to feedback into the input, to create a thinking loop, and they're optimizing that.

Same here, I tried many models GPT-5.6 Luna/Terra/Sol, Claude 5+, DeepSeek, Kimi ... at last I came back to Claude 4.6 that feels so much simpler and smarter than all new models.

We moved from Gemini to Claude recently, and while I absolutely love Claude's coding capabilities, I often find myself extremely overwhelmed whenever I ask Claude to explain something to me, whereas that was a breeze with Gemini Pro models.

It generates pages for what could be explained in 1-2 paragraphs.

I use to love using Gemini CLI to have Gemini explain a module for me before I dive into the code and point out specific points of interest, but I can't do that with Claude Code without feeling overwhelmed and starting to heavily procrastinate

Reply on news.ycombinator.com