OpenAI Model 54% More Efficient for Coding Agents - AI is Now Only Stupidly Expensive

54% more efficient translation: Instead of falling from 80th floor you'll be falling from 40th floor, much better deal according to OpenAI

The problem is diminishing returns. A nonreasoning model gives you a decent answer with just a handful of tokens. A reasoning model gives you a somewhat better answer by burning a lot more tokens. And an agent gives you a slightly better answer by setting an entire token budget on fire. The catch? Output quality grows more or less linearly, while compute and token consumption grow exponentially. So they've finally hit the wall they kept insisting didn't exist. It's called cost. Sure, with infinite compute you can achieve almost anything, just keep rolling the dice until you accidentally stumble upon the right answer. We knew that long before AI. It's called brute force, not intelligence and not AGI.

Tokens are cheaper so now developers can now generate more cat photoes.

Damn Eli doesn't take any days off..... I love it! 😆

if they make the service "cheaper" they are knee-capping their future profitability. it has become a self-fulfilling prophecy atp.

Well I'm glad someone is talking about it!

You, Ed Zitron, Yann Lecun, Gary Marcus, Cal Newport, etc will all be remembered very fondly once this is all over. People who know what they are talking about and have no financial incentive to lie about AI. Those are the ones that will tell you what's really going on. These idiot investors trust these conmen CEOs. We'll see who gets the last laugh.

Lamborghini cheaper if you buy 10 instead of one

123, remember this number, this is when elon ceases to be a trillionaire :D

Good thing energy and computer components are only getting exponentially cheaper for the foreseeable future! Two inches forward for efficiency, six mile backwards for everything else.