GPT 5.4 vs 5.3-Codex
- 内容介绍
- 文章标签
- 相关推荐
Hello everyone!
I was just wondering what your experiences were with GPT-5.4 versus GPT-5.3-Codex. It says on benchmarks, of course, that the newer 5.4 is better, but this hasn’t been true in my case, as I get the sense that 5.4 isn’t aligned with me on some projects while 5.3, while warning me, still goes ahead with stuff anyways. And the the ‘less token usage’ of 5.4 I didn’t see play out, as my limits haven’t been reached at all
To me, 5.3 is just better at getting stuff done. Maybe 5.4 is better for math and visual reasoning, but 5.3-Codex is my preferred. What are your thoughts?
image1194×152 14.9 KB
网友解答：--【壹】--：
I haven’t used 5.3-codex very much, but IMO the newer model doesn’t mean better performance in all aspects. If you find some tasks that 5.3-codex does better, just trust your experience.
--【贰】--：
一般情况还是用5.4吧，除非是纯编码任务，可以派个5.3codex子代理去做 ，但是现在5.4mini又快又便宜，更多的时候不怎么用5.3codex了
--【叁】--：
真的可以在L站看到外国的佬
Wow, I didn’t expect to actually see foreign bros here.
--【肆】--：
--【伍】--：
We are banned by Openai
--【陆】--： Amsen:
We are banned by Openai
We are banned by Openai 没毛病
--【柒】--：
Bros, can someone explain the ‘we are banned by OpenAI’ threads? I’m a bit confused and haven’t been reading recent posts
--【捌】--：
We are banned by Openai
--【玖】--：
That’s interesting, to my experience 5.4 was less good at following vague directions than 5.3-Codex-High. 5.3 “one-shot” some projects, so to speak, while 5.4 needed finer tuning, sometimes even causing me to roll back changes in the git repository.
--【拾】--：
5.4-mini 相比5.4 会弱多少？
--【拾壹】--：
就使用体验来讲，编码能力双方不相上下，但是项目组织能力，理解能力交互能力，还是gpt5.4远大于codex模型
如果你有一份很详细的文档，codex是没问题可以胜任的，但是只有一些模糊的需求，那么codex比不过gpt5.4
--【拾贰】--：
写代码用5.3-codex
--【拾叁】--：
We are banned by Openai
--【拾肆】--：
We are banned by Openai
--【拾伍】--：
作为非程序员，我还是喜欢gpt-5.4，因为它的回答我能读懂，但是5.3-codex经常读不懂。但我也看了评测，代码能力5.3-codex更好，所以我一般配置写深入代码的subagent用5.3-codex。
顺便， ‘we are banned by OpenAI’是因为openai禁止中国大陆地区访问，我们只能想一些办法，然后最近openai封禁了一大批账号……
--【拾陆】--：
We are banned by Openai
--【拾柒】--：
We are banned by Openai
--【拾捌】--：
我更喜欢gpt5.4，更加喜欢新的模型，普通简单的任务使用高，复杂的使用极高推理等级
--【拾玖】--： Amsen:
We are banned by Openai
We are banned by Openai
Hello everyone!
I was just wondering what your experiences were with GPT-5.4 versus GPT-5.3-Codex. It says on benchmarks, of course, that the newer 5.4 is better, but this hasn’t been true in my case, as I get the sense that 5.4 isn’t aligned with me on some projects while 5.3, while warning me, still goes ahead with stuff anyways. And the the ‘less token usage’ of 5.4 I didn’t see play out, as my limits haven’t been reached at all
To me, 5.3 is just better at getting stuff done. Maybe 5.4 is better for math and visual reasoning, but 5.3-Codex is my preferred. What are your thoughts?
image1194×152 14.9 KB
网友解答：--【壹】--：
I haven’t used 5.3-codex very much, but IMO the newer model doesn’t mean better performance in all aspects. If you find some tasks that 5.3-codex does better, just trust your experience.
--【贰】--：
一般情况还是用5.4吧，除非是纯编码任务，可以派个5.3codex子代理去做 ，但是现在5.4mini又快又便宜，更多的时候不怎么用5.3codex了
--【叁】--：
真的可以在L站看到外国的佬
Wow, I didn’t expect to actually see foreign bros here.
--【肆】--：
--【伍】--：
We are banned by Openai
--【陆】--： Amsen:
We are banned by Openai
We are banned by Openai 没毛病
--【柒】--：
Bros, can someone explain the ‘we are banned by OpenAI’ threads? I’m a bit confused and haven’t been reading recent posts
--【捌】--：
We are banned by Openai
--【玖】--：
That’s interesting, to my experience 5.4 was less good at following vague directions than 5.3-Codex-High. 5.3 “one-shot” some projects, so to speak, while 5.4 needed finer tuning, sometimes even causing me to roll back changes in the git repository.
--【拾】--：
5.4-mini 相比5.4 会弱多少？
--【拾壹】--：
就使用体验来讲，编码能力双方不相上下，但是项目组织能力，理解能力交互能力，还是gpt5.4远大于codex模型
如果你有一份很详细的文档，codex是没问题可以胜任的，但是只有一些模糊的需求，那么codex比不过gpt5.4
--【拾贰】--：
写代码用5.3-codex
--【拾叁】--：
We are banned by Openai
--【拾肆】--：
We are banned by Openai
--【拾伍】--：
作为非程序员，我还是喜欢gpt-5.4，因为它的回答我能读懂，但是5.3-codex经常读不懂。但我也看了评测，代码能力5.3-codex更好，所以我一般配置写深入代码的subagent用5.3-codex。
顺便， ‘we are banned by OpenAI’是因为openai禁止中国大陆地区访问，我们只能想一些办法，然后最近openai封禁了一大批账号……
--【拾陆】--：
We are banned by Openai
--【拾柒】--：
We are banned by Openai
--【拾捌】--：
我更喜欢gpt5.4，更加喜欢新的模型，普通简单的任务使用高，复杂的使用极高推理等级
--【拾玖】--： Amsen:
We are banned by Openai
We are banned by Openai