知Yizhi Daily
← Back to this edition

Conversations

V2EX user compares Claude and GPT: trade-offs in stability on complex tasks

Edition 2026-11-21 · 08:00 Beijing time

Based on project experience, a V2EX user says the online claim that Claude is overrated has some truth: some model versions produce bugs, and its frontend aesthetics did not meet expectations; GPT's foundational abilities are not weak, but its style is rougher. The poster appreciates Claude Code's confirmation and commit workflow and asks what everyone mainly uses day to day.

The poster says that after continuously using several models on a project recently, they feel the claim that “Claude is overwhelmingly strong” comes with a filter, and in many scenarios it is even not as good as GPT. They mention that using 6 astra and 6.1 sol in the project was mostly acceptable, but switching to opus5.5/fable 5.2 occasionally produced bugs that had to be fixed afterward; the new 6.2 sol also disappointed them because it stopped halfway through implementing a feature.

On frontend work and aesthetics, the poster says they had heard Claude was strong at writing frontends, but after using it to rebuild the official website, the resulting black, brown, and green color scheme was not striking and was even inferior to a previous version made with 5.6 and then gradually polished with 6 astra. They assess GPT's copywriting and code style as often rather “rough,” but its underlying capabilities are not weak.

At the interaction level, the poster considers Claude Code quite pleasant: after completing each segment of work, it presents an interaction for confirmation or commit, and after clicking they can continue moving forward; they say they have seen this kind of experience less often in Codex.

The poster summarizes their feelings with an analogy: Claude follows a more refined route, similar to Apple; GPT follows a heavier, hard-core route, similar to Google. But on complex tasks and stability, they think Claude is not necessarily as strong as public opinion suggests, so they ask other users which company's model they primarily use day to day.

Original sources

V2EX ↗
Back to this editionOpen interactive reader