It happened. Claude Code has been dethroned. I've lived in Codex w/ ChatGPT 5.5 the last couple of days and it is the: • Smartest model • Best all around AI app • Most powerful features (computer use and image gen) Here's everything you need to know about this... A+ release:show more

Love it! Staying on Claude Code opus 4.7. Been building for a month and I’d really need to be convinced to make a move for anything more than 25% better. Instead of risking losing momentum to learn another coding tool. Or, am I being shortsighted? #finnfam

you're allowed to use whatever you're comfortable with but it's unbelievably easy switching tools. there's literally 0 switching cost. you just hit new session in the other tool and it works.

Dude you are so freaking good at this

easy to make content when you love what you're talking about

@LLMJunky Permanent Upper class

@AlexFinn i'm relegated to the shadow realm

Why chose? Using both is the unlock.

been switching between both this week. codex is genuinely impressive but claude code still wins for me on long agentic runs. anyone else finding the same?

Great video. One reqesut. Can we please stop saying "complete game changer" every time a new model comes out? Lkl

Are all the long forms of these videos on yt????

my yt exclusive content is the live streams. i live stream there every monday wednesday friday at 11am pst

"Dethroned" feels strong after a couple of days. Models trade leads every other release cycle now — Codex/5.5 looks great for some things, Claude still wins others (long-context refactors, IMO). The honest move is a hot-swap setup, not loyalty to one lab.

For greenfield tasks and computer use, agree — Codex + GPT 5.5 is incredible right now. Where Claude still wins: deep refactors with long context and codebase memory. Both tools, context-switched. That's where I've landed.

I've given 5.5 a big test with legal reasoning work. Very good, but I'm sticking with 4.7 for now. It just feels better to me. And, I suspect we can safely assume that opus 4.8 or 4.whatever will be along soon in this great game of AI leapfrog.

I just made the switch to, Alex

the smartest model isn't the bottleneck for me on a 200k line codebase. claude code still wins on filesystem discipline and the skills/mcp surface. codex feels like a brilliant intern that forgets where files live. on greenfield it's a different story tho.

he's finally joined us

same

Why does 5.5 feels the same like 5.4 for me in terms of feeling like you’re talking to a corporate bot? Opus is so great with reasoning, understanding the business logic of the app and taking it in consideration with every brainstorming about new features.

*cough* Bullsh%t. Google + Anthropic FTW. Not taking away from their shine, or the intelligence of their 5.5 model, but Bullsh%t, 15 minutes from now it's all new and different again, and 15 min again, and so on, and lands with G&A, and soon xAI. #facts

All I know is that it has been proven to have very left wing stance when it comes to facts.

the benchmark wars never end

3 days of usage and the verdict is in. give it 6 weeks then we'll see

Sucks so far for me, burned through my allotment with 2 prompts.

We run 12 AI agents in production. Some on Claude, some on GPT-5.5. "Dethroned" is the wrong frame. If swapping models isn't a config change in your stack, the model isn't your problem - your architecture is.

The designer in Claude works similarly, allowing you to reduce and refine in several ways until you reach the desired look.

yeah unfortunately codex is killing claude code on all fronts besides design

Codex with 5.5 is actually insane for building stuff fast. The way it controls the desktop and generates UI images on the fly is next level. Claude is still better for huge refactors though. Feels like I’m gonna end up using both depending on the job. Solid demo man

nolan codes in claude code every day and hasn't so much as glanced at codex. these dethroning posts always seem to come from 48-hour tests, not from anyone whose actual workflow has moved. switching cost is doing the real arguing.

With all respect, what you have said feels like paid promotion for codex. The browser use is there in claude code for ages, just say "use /chrome to test your changes" and see the magic. Linear integration also has nothing to do with codex and was working fine with cc and curso

@soumitrashukla9 Building in Codex is amazing

gpt 5 still hallucinates on complex builds. told me it completed 2 modules. checked the codebase. they didn't exist. 12 module project, 2 phantom completions. not touching it for anything beyond prototyping

We have created the open source version of this. You should try out Thoth. Its does pretty much what openclaw/codex do but is so much more user friendly. It even has a built in designer studio now. Its quickly becoming AI Super app people have been talking about.

dude Alex you're so clickbait ... meanwhile I just built a calendar app, check it out you won't regret it

Claude be like: “My comeback will be legendary”

been testing 5.5 pro for automation QA the last 48 hours it's the first model that catches edge cases i didn't think to test for... like it actually understands what "this will break in production" means

🇧🇷🇧🇷🇧🇷🇧🇷👏🏻 it's fast, and codex is amazing!

we called it

raw IQ isn't my bottleneck on a 200k line repo, harness discipline is. codex still loses 30% of context on long sessions vs claude code's agents.md cascading. the model leaderboard and the shipping leaderboard aren't the same thing.