TBH this is such a pity - 12 months ago I would have bet everything on them. They are the only one of the BIG players who own Model + Infra + Reach/Adoption. They are the only ones who could take the Apple approach; design TPUs/Infra specific optimized for their hyperspecific usecase, and dominate the market... such a pity
@F1 & @Hasbro Your newsletter signup for the F1 Monopoly Game release.. doesnt work? There is no button, no box to tick, no terms to review. Nothing. I dont care about F1 Financial News (oddly placed given this page is for a board game release). Also, "350 other subscribers" is hardcoded onto the page. Did yall vibecode it? LOL
Usage resets are nice to have, but we need more stable connection and a "fast" mode that actually is "fast-er". Right now, there is no difference between the two, and keep getting constant reconnecting. What is happening? @thsottiaux
@emanueledpt You need to reasses the scope - what is even the purpose of remodex? Not trying to be a dick about it - genuinely wonder if this was a tiny passion project that turned into more (faster than you could answer the "why") or if it was meant from the getgo as a potential business
@theo The Codex App has completely replaced my terminal - especially for UI work. The only case where the terminal is still my goto choice is for adhoc usecases like remoting into a VPS etc.
(I dont remember exactly the current situation on the iOS App Store revenue split so dont quote me on this)
Could it be that, on Mac, you login to chatgpt web, and then download the DMG to install Codex App. On iOS, they would need to release it on the app store, and they (I believe) cant ask users to sub from outside the app. This would mean that Apple would get 30% of the subscription value.
@thsottiaux hey man, since the last update, when I ask gpt 5.5 to spawn subagents, codex ends up using 120+gbs (M4 pro, 24gb ram).
Had to a take a picture with my phone because the laptop was dead frozen. This was with 4 or 5 subagents running just exploration in a small codebase
Im confused - @cursor_ai was in talks to (maybe) be bought by @SpaceX (Cursor gets compute, Elon gets a home for Grok). But then they hand out compute to @AnthropicAI (money now-'ish' is better than money tomorrow-'ish'). So what is gonna happen with @cursor_ai ?
Only me or Fast is actually slower than Default for Medium thinking on GPT 5.5? @thsottiaux what's happening? Between all the "reconnecting" and overall poor performance, Standard is now faster than Fast.
Burned through 40% of the weekly quota in less than 2 days. Exta High + Fast is not sustainable with the current Pro Plan limits. Then again, I am surprised by how fast the non-fast mode can be at times, and how smart Medium is compared to Extra-High. EH still wins, but not by a lot
Re Claude being an unpleasant experience for average users due to the model selector - my dad (example user) has no clue of how much models actually cost. Already paying 20$ (for something he has for free for years with ChatGPT) means he expects "more intelligence". In Claude, that means selecting Opus, but that will kill his weekly limit in 2 prompts, make him cancel his Claude $20 p/month sub, and return to OAI.
The problem is not Opus' pricing, but the option to select it to begin with.
C, but I also think that the average user of the ChatGPT app, and the average user of the Codex app, are 2 completely different users. One is only looking for answers to questions (no need for even things as simple as a model selector - IMO what causes the biggest issues with new Claude users), the other wants enough pro-features to have Codex become their superapp, but not too much customizability like PI would allow you.