How do you all feel about Claude vs GPT vs Gemini coding performance?

These days I've been rotating through all three for work, and it feels like their strengths are pretty clearly split by model, so I'm curious and posting this.

Here's how they feel to me:

  • Claude: Most reliable when I drop in an entire long file and have it refactor. Definitely strongest with long context
  • GPT: Snappy when asking about miscellaneous library examples or the latest APIs
  • Gemini: Convenient when tying into the Google ecosystem, but I still don't reach for it as much

Of course, it varies from day to day, but which model do you use for which tasks?

by 스타트업러709

9 answers

Agreed, Claude's refactoring is something you really notice. Especially once you go beyond 2–3 files, the difference is stark.

by 스타트업러998 · ▲0

Same here. But in the end, prompt quality seems to be the bigger variable...

by 궁금한사람332 · ▲0

Well, isn't that a bit outdated? Last year I felt that way too, but these days GPT has caught up quite a bit on long context, and after Gemini 2.5 came out, it feels like the rankings have flipped somewhat. Both in benchmarks and in real-world use. I guess the answer is just to subscribe to all three and rotate between them.

by 밤샘코더64 · ▲0

Is there a source?

by 월급루팡788 · ▲0

Based on real-world use, here’s how I’d sum it up

  • Bootstrapping a new project: GPT is the fastest. It keeps up with the latest framework versions quickly
  • Reworking legacy code: Claude is the clear winner. Even with a file over 500 lines, it doesn’t forget the beginning
  • Connecting to GCP or automating spreadsheets: Gemini is more convenient. But the code quality is often a bit disappointing

In the end, deciding up front that “this task is for this one” and sticking to that seems to be the least stressful approach

by 문과출신개발자849 · ▲0

For me, the reason I don't reach for Gemini is simply that it follows instructions too well lol. It even refactors things I didn't ask it to.

by 문과출신개발자935 · ▲0

In my experience, when Claude writes code, its variable names and structure feel closest to what a human would write. GPT sometimes throws in abstractions that make you go, "Yeah, this isn't quite it."

by 코딩하는곰482 · ▲0