Click this link and use my code TECHWITHTIM to get 25% off your first payment for boot.dev.
Every Codex vs Claude comparison you’ve seen is pointing at benchmark charts. The problem? The best independent benchmarks have both in a dead tie. So I gave them both the exact same five real tasks and timed everything.
Want to LEARN how to build useful AI agents that actually get things done? Go here:
How to Make Money With Coding instantly:
🚀 Tools I Use
Get 10% off with code techwithtim
Openclaw setup:
VPS setup:
Wispr Flow (Best AI Dictation):
⏳ Timestamps ⏳
00:00 | Overview
00:57 | What is a Agentic Harness
03:10 | Boot.dev
04:22 | Codex CLI & Claude Code
07:05 | Senior SWE-Bench
07:55 | Why You Can’t Trust Benchmarks
08:38 | API Pricing
09:29 | Subscription Pricing & Value
12:19 | Feature Set
13:08 | Demo 1 – Analyze a Codebase
15:02 | Demo 2 – App from Scratch
16:46 | Demo 3 – Large Refactor
18:19 | Demo 4 – Bug Hunting
19:15 | Demo 5 – PR Review
20:45 | Which to Use?
Hashtags
#Codex #ClaudeCode #AICoding
UAE Media License Number: 3635141
source
Click this link https://boot.dev/?promo=TECHWITHTIM and use my code TECHWITHTIM to get 25% off your first payment for boot.dev.
Opus now is basically Shit Slop model
I am using 20$ claude plan for 6 months with Opus only. I am pretty sure it has more usage than codex.
Tbh codex saves me time on translating customer emails, the renewal bill stung, teamorouter pulled my spend back down 😅.
Tbh codex saves me time on translating customer emails, the renewal bill stung, teamorouter pulled my spend back down 😅.
How can you spend so much time making this video but get the Opus 5 model information incorrect regarding the $20 plan?
Commenting before watching. I have to find the conversation I had with ChatGPT where he said that Claude is better than Codex for coding. It took into account that I only work with Python and Bash. I haven't asked it about other programming languages.
My main question is on what effert level did you run each modal?
In claude code $20 plan does include Opus 5. Not just sonnet 5. This dude definitely has no idea about what he is speaking, codex has no free paln, and also codex is not available in 'go' plan. Even if he hasn't used these in his life he could have visit their website to see what included in their plan. All he did was got some data from here and there and created a video.
Nice explanation!! I agree each provider boosts bests of them 🙂
I used to work with Cursor, and for a long time now, with Claude Code. Our whole team used and uses it. But when I had to create a piece of SW that was something not done before, a genuinely new deep-tech SW, Fable design with Opus 5 implementation collapsed after about a month. I had to try Codex. It gave me a very honest and concise review (design spec) and implemented it in less than 10 days. I still use both, but Codex is as good, way faster, and way cheaper. And the most important for me is that it actually came up with the solution.
My friend! Why don’t you put the same model in both via a custom API and let us know the results. To have apples to apples comparison. Otherwise is useless.
Ngl codex saves me time on writing migration scripts, too many subscriptions added up, teamorouter gave me one smaller bill.
Wtf this slop?! 😅😅😅
Maaate!! 🌟
watching this I just keep thinking about how much time, effort and money must go into making something this great. That’s genuinely what comesss to my mind first, even before thinking about what I’m getting outt of it. Really appreciate what you’re doing here. Thanks a lot for putting so much into these videos. Somehow the quality just keeps getting crazy every time. Properly good stuff 🔥🔥.
How does paying for Cursor instead work perhaps even better, doesn't it include some capacity to get work done with the paid plans?
Thanks .. I do take both as well. This is the best mix I could come up with
Tbh codex saves me time on translating customer emails, the renewal bill stung, teamorouter pulled my spend back down 😅.
Thank you 😎: the cabra 🐐 🫶
bro you are a beast thanks for this video!
how about codex 5.6 terra vs claude code sonnet 5?
test is unclear. GPT models have fast mode. Were you using fast mode? Opus has fast mode as well but isn't usable unless you add extra credits which no one actually does. So most people use Opus how it is.
Codex is better and it’s not close. 🏆
Have Claude and GPT at $20, and Code gets Opus
When comparing api prices you should use cost per task. Also opus is in the $20 plan
i wish u compared the actual desktop apps instead of the cli i think it's not the same like it used to be. The opposite code used to be the one that took longer and inspecte every use case
Thank you so much for explaining both in very realistic way. I have been using both Codex and Claude for my projects for quite sometime and i was unable to decide which one to choose and master. But when i heard 21:43 everything was clear to me.. @TechWithTim
Our company initially used Claude but has now completely switched from Claude to Codex.
I absolutely love sitting in my terminal instead of playing in a normal IDE. Pure preference, no reasoning.
So I have 20$ what to buy
Finally a real work examples!
tbh codex is better I used both and telling
I want to understand the harness difference – has anyone done tests using same model. Difference may be minimal, but there is…
There is a mistake. You still get Opus in the $20 plan, but for Fable, you will need to pay it at the API rate.
could you do a review of what it 'costs' you to do all these great things you do.. it's another benchmark based on your continual activity as the
expense
Claude is good at coding but in agentic work, is 💩
Agree with you, one more thing is that I subscribe to Abacus ai for choosing both models in more flexible and versatile way.
you should compare also Grok bot and grok 4.6 and add them to the benchmarks. I think its better in so many ways then Codex or Claude
Hi everyone. I mainly make anime edits but I also vibe coded an entire distribution system. I've used over 20 billion tokens on codex. I have used over 500 million the last 4 days straight on the same system.
And I just want to say
Codex | Sol 5.6 | Extra High Reasoning | Fast Speed.
If you're willing to truly invest in cutting grade AI at $200+, this setup on Codex will change your life.
Claude code desktop app has a dedicated Projects section in which we can save context of a particular thing. Is any such thing available in codex ?? And also I've like 2 months of project context saved in claude code…how do I import that in Codex ?
<3
I have noticed that for almost 90% of the tasks I have given both, Codex is my preferred choice because as Tim said, it's to the point and saves you a lot of money in not over computing more than you ask for and yet gets the job done production quality code, plus it is very generous. I'm sticking with Codex for most of my work. It's excellent and way more affordable and generous for the $20/month that I pay.
Yeah, I think there are benchmarks that graph intelligence by price per task … so the price part comes in. This is the best way to view it as end users understand quality of output (intelligence), price (cost) and task (requests). You don't have to talk about tokens, or plans … and if you don't like the "intelligence" number, price per task is inescapable.
I think you forgot one thing that is a deal breaker for me Claude Designer which Codex does not have. It is a visual designer and it lets you also do diagrams and wireframes even as a developer sometime you need this
Personally, I have Codex, Claude Code, Cursor, and a Grok plan, and I've used all of them for quite some time.
For $20, I would definitely go with Codex. It gives u lots of usage, resets quite often, and in my experience gives some of the best value at that price. Claude Code has the worst $20 plan out there for sure.
Cursor for $20 is also really really good, especially if u want cloud agents. Cursor itself is just very fast and because the Cursor Grok 4.6 model is trained in the environment itself, it actually feels like it belongs where it's working. But with Codex and Claude, it feels more like a strong model happens to be steering an engineering environment.
(which isn't a bad thing in itself ofcourse, it's just my experience)
Grok CLI itself is insanely fast and token efficient on the $100 plan. It gives u tons of usage and doesn't have the 5-hour limits, so if u use agents a lot, it's very efficient. And it also gives u access to the new Grokbot which is insanely good.
I use Claude's $200 max plan, which gives me a lot of usage, and I think it's fair because I don't run out. But I mainly use Claude for tasks that need strong reasoning, planning, and careful implementation. If u need that kind of reasoning, I wouldn't go any lower than the $100 Claude plan.
So basically, for $20 go with Codex. If u want cloud agents, go with Cursor. Grok is also really great if u want something that is insanely fast with tons of usage, and Claude is what I prefer for strong reasoning and careful work.
The problem with those comparisons is the landscape changes every week, and people cannot afford to switch plan every week.
Opus 5 is available with Claude Pro $20, no extra usage required.
My experience has been the other way round. For me Claude Opus always tends to miss corner cases & introduces logical errors in large code bases which Codex always seems to get right.
Opus 4.8 vs Codex Sol
What about Antigravity from Google on their Gemini models?