跳过导航
Claude Sonnet 4.6 vs GPT 5.3 Codex
0:00 / 0:36

Claude Sonnet 4.6 vs GPT 5.3 Codex

1LittleCoder💻

@1littlecoder2.2万 位关注者

在 X 上查看
760 个赞下载原帖

142,985 次观看

Claude Sonnet 4.6 vs GPT 5.3 Codex Same Prompt in Cursor (Neutral Harness) - Default Settings! GPT 5.3 Codex took - 2 mins Claude Sonnet 4.6 - 7 mins Cheaper than Sonnet 4.6! Faster than Sonnet 4.6! But I'm not sure if it's more craftier than Sonnet 4.6 because Sonnet's got a taste! Claude Sonnet 4.6 had a better result (aesthetically) and also more solid techstack - GPT 5.3 codes meanwhile just got the job done!

36 条评论

评论来自 X (Twitter) 上的原帖

  • Barrak Ali | براك علي 的头像

    @BarrakAli

    Speed vs craft is always the real trade-off. Sonnet 4.6 taking longer but potentially producing more refined output is exactly the kind of nuance that matters for real production use cases.

    13 个赞

  • Jean-Philippe Borel 的头像

    @BorelJP

    If you want to build a solid app with a solid architecture and a cleaner codebase you go for Sonnet 4.6. If you want to build something quick and dirty Codex. Is that the conclusion?

    5 个赞

  • Mike 的头像

    @WorldsByMike

    I love starting a project with Gemini 3 pro in antigravity and then switching to codex. Works quite well for front end and then codex for the rest.

    4 个赞

  • Ravi 🇸🇪 🇨🇭 的头像

    @Lecstappilton

    Unless your dev cycles come from 2 weeks (min) to 2 seconds, nobody cares who slow a tool is when it comes to hours.

    3 个赞

  • Anayat 的头像

    @anayatkhan09

    Neutral harness is a nice sanity check but in real codebases the big swing is how many backtracks and tool-call retries each model needs. Curious if you have traces for full tasks, not just one prompt in Cursor.

    2 个赞

  • Robert Hu 🦉 的头像

    @theroberthu

    Speed gets it done, taste makes people actually want to use it

    2 个赞

  • Toni Hintikka 的头像

    @tonihintikka

    Codex is more comparable to Opus.

    2 个赞

  • 1LittleCoder💻 的头像

    @1littlecoder

    pricing?

    1 个赞

  • John Dinsdale 的头像

    @JohnDinsdale48

    why is speed a measure?

    2 个赞

  • 1LittleCoder💻 的头像

    @1littlecoder

    more important for agentic use-cases and also obviously the time in which it completes a project matters a lot for iterations

  • James Cooper 的头像

    @vaultminds

    Seems to be the theme with Claude, has always been better at design and creative things👌

    1 个赞

  • Sanchen007 的头像

    @mimighost008

    You can have GPT 5.3 code it up and sonnet to optimize the interface loll

    1 个赞

  • JK 的头像

    @_junaidkhalid1

    My experience: the “got the job done” outputs usually come back to bite on the second iteration. I end up spending time untangling weird abstractions or missing edge cases. For quick prototypes, I’ll take the faster/cheaper model every time. For anything that needs to be maintained, I’d rather pay more if it saves me 30 minutes of cleanup and makes the codebase feel coherent.

    1 个赞

  • Rita 的头像

    @Rita53135643

    Efficiency vs. Elegance. 🥊 GPT 5.3 Codex is a beast for raw speed, but Sonnet 4.6 proves that "vibes" in coding are real. Codex builds the engine, but Claude designs the Ferrari. 🏎️✨ > When the stack is solid and the UI has "soul," those extra 5 minutes are a rounding error. Are we coding for benchmarks or for humans? Team Claude for the win. 💜🦾 #Claude46 #GPT53 #CursorAI #Coding

    1 个赞

  • Visionati 的头像

    @visionati

    This matches what we see at Visionati. We fan out to both Claude and OpenAI (plus Gemini, Grok and others) for image analysis. Different models consistently notice different details in the same image. Speed vs depth is a real trade-off, not just for code.

    1 个赞

  • Bilal Bakr 的头像

    @bil0090

    Codex is a model thats made for ppl who are in tech, its like a dev, needs details & orders

    1 个赞

  • Shubh Jain 的头像

    @shubh19

    i think the main difference comes down to tone and nuance some tasks just require a bit of swagger and flair regardless of efficiency or speed sonnet 4 6 has that certain je ne sais quoi where gpt 5 3 is more like a utilitarian tool

  • Abu Bakar Siddik 的头像

    @abubakar_AIE

    Everything is a trade off in this space

  • Ashish Patel 的头像

    @ashishxcode

    Speed vs craftiness is the real dilemma. Sonnet's reasoning often feels more intuitive for complex tasks, even if it's slower. GPT 5.3 Codex is a speed demon though. Which one actually got the logic right in the end?

  • Twlvone 的头像

    @twlvone

    speed is the wrong metric. the question is how many iterations does it take to get to done. a model that takes 7 mins but nails it first pass beats a 2-min model that needs 4 rounds of fixes every time. total time to correct output is the only number that matters.