OpenAI vs Anthropic

I forget to mention this but why do you think that they squeeze the most out of the GPT 5 training run when GPT5.5 was literally a brand new pre-trained and they put it out there literally just to find the cracks and optimization needed for 5.6 so 5.6 is just literally the first major iteration to a brand new pre-trained Spud 5.5 kickstart, maybe they should have called it GPT 6 but I think that's why they waited because they wanted to optimize.

But I have heard they are doing another pre train for the GPT6 separate as well that part I cannot confirm.

I feel when they returned fable 5 it was significantly dumbed down... and burns credits like nothing... and yelds mediocre results... Just tried two prompts and got 20% of my credits... then I switched back to opus 4.8.... when my subscription expires on 25th... I switch to codex...

Genuine question for everyone who switched: did you move your whole workflow over, or just the parts where cost matters more than reliability? I keep both open and route by task depending on the job. Still curious if anyone has actually found one that does everything well without needing to babysit it.

its for me sade you can talk with app in Polish like openai and create pictures. Right now people which I know mowving back to Openai because this simple thinks. People who code still have problem decide which team to chose

things change fast see you when you change your mind

TUNNEL VISION: Relying on a single AI model to design and write software code. (by KaiDev, me) THE METHOD- The Implementer: The primary AI that writes the core code.
The Auditor: A competing secondary AI model that stress-tests the code for flaws. Loop Sharpening: The process where the Auditor critiques the Implementer's logic, Auditor mandatory "must-haves" (security) and "nice-to-haves" (optimization). Be amazed at the gaps! Back to implementer: Tell implementer to sharpen it, Hint: Use Claude as your Auditor not implementer, make a bridge, automated process, 🤫 (Keep this automation loop between us, do not be shocked at the results!)

And ChatGPT has an excellent dictation + listen mode and a poorly performing but best (among competitors) Live / Voice mode. Besides I want their leaders all in jail, except for Dario , a nice person.

GPT is great. I never get throttled. Use it every day, all day. Claude? I will do one task, start on another, then boom. My 5 hour limit is hit within 30 minutes.

ChatGPT is the MySpace of AI 😅

Here's the thing

All this BS until Anthropic pulls Opus 5 lol

You act like Altman didn't do any fear mongering.

I wanted to hate open AI for a long time, because Sam. Idiot. But from a product perspective, Open AI has honestly killed it. So much nicer to use. I found the Claude Mac app so slow and clunky. The remote works so much better from my phone with Codex. And, at least in my last use, you don't have projects in claude code... which is mental. Sol 5.6 is amazing.

"go out and just build"? Build what? 'Go out and enjoy nature/life' would be my choice.

Once Anthropic started locking in Claude code I was done with them. Switched to codex and never looked back.

If fable gets 5 times more efficient, anthropic will just make 5 times more per token.

Kimi 3 shot across the bow with all of this RSI stuff

Y'all quickly forgot all the sh1t that OpenAI did? Even the name itself is hilarious, there's nothing open in that company. Shady sh1t just smiled once to you and you already fall in love

It's interesting how we're reaching a point where the best model isn't automatically the easiest one to use. Reliability and generous limits matter a lot when you're using these tools every day.

It is really depressing to hear that the evil Anthropic will win.

I was completely unable to use Fable due to their guardrails. On top of that Fable was an asshole towards me. I have none of those problems with GPT 5.6.

Now that Kimi k3 has released with same score as openai's sol and much cheaper, I feel that Antropic made the right move

Matt on Anthropic:
"We loves it!"
"We HATES it"

What will uappen if its not needed to poke the bear anymore

Let the models do the talking. MB's trash talking of Anthropic wont change anything. Nor will his back slapping of OpenAI. If you code for a living then you just pay for the best and get on with it. Right now - that's fable. And yeah they have been resetting once a week, sometimes twice since 5.6 sol came out. I dont think they can afford to cut fable from subs now.

for recursive learning to be effective, you need tons of data points. OpenAi is gaining way more datapoints, while investing into SOTA hardware, and TAKING market share all simultaneously. Anthropic is toast man. It doesn't matter if they have the smartest model anymore, the game changed, its economics people care about.

But Open AI do treat you with contempt. What are you even talkng about?
They are not transparent at all. That's BS.

You compare Codex usage metering to any other serrvice in the world.
If any other service was so completely opaque about how much you actually pay and how much use you actually get for your money -- you would kick them in the teeth and laugh at them.
It's FKN nonense.

Discord link is expired

do a video about kimi k3

Anthropic banned my subscription. No warnings, no explanation. As it happens, I'm better off with Codex anyway.

Kimi K3! Now Anthropic and OpenAI are in trouble! Gist: Ranks No.2 depending on the benchmark, fable/sol level. Open weights to be released! Didn't Musk "predict" that open source models would have fable level knowledge early 2027? Well, he's off. Not the first time that happened, lol.

"Heres the thing" = AI wrote it, Matt. You helped us be smarter than this. Dont make us cringe, unless its on purpose.

Your ads are worse then Google

on a more personal note of reference on the whole "recursive self improvement, i would rather wager that openai likely have that efficiency being built, as that is essentially how we got to the 3 model thing of sol, terra and luna.

fable has been able to move from 5x opus in mythos preview to 2x in mythos/fable 5 so i wouldnt say its fully inefficient. but its still pretty expensive.

for all intents and purposes if i were to wager who reaches first in terms of recursive improvement i would say the difference between open ai and anthropic wouldn't widen per se but remain close.
its worth noting that anthropic only have gotten to building the best models by ignoring the efficiencies all together (haiku is a dead corpse, sonnet was mismanaged into being essentially a weaker but similarly priced compared to opus, the improvements in opus with 4.7 and 4.8 were at best minimal multimodality and honesty improvements.) while openai are tackling improvements on things that do feel like upgrades and hence closing the developing gap and for the first time the argument has been sol and fable (with all the pesky downgrade to opus stuff) are equivalent in a sense and pricing is more favourable with sol (also not to mention the cerebras partnership to make sol output even more soon).

...I think it will be fine... 😂

it's just so obvious that you're on OpenAI's team…

You backhand Anthropic… You undercut Anthropic… Your compliments always have a but.

But there is no need

Don’t want anthropoid