GPT-5 optimizes costs

The Register reveals OpenAI’s strategy and according to them, GPT-5 turned out to be not a revolution of capabilities, but genius cost optimization.

Instead of a monolithic model — composition of at least 2 systems: light and heavy, plus intelligent router. Imagine — each request is analyzed and the system automatically chooses the optimal model. Simple question — light model. Complex task — heavy artillery kicks in. Computing savings are enormous!

Automatic reasoning management becomes a key tool. Reasoning is activated only when truly necessary. Free users can’t control this. Less computation, fewer tokens, radical cost reduction. The smart system decides itself when deep thinking is needed and when surface answers suffice.

But why so? 700 million active users per week, but only 3% paying! ChatGPT became synonymous with AI, like Google became synonymous with search. But such leadership requires astronomical infrastructure costs.

Strategic limitations work for optimization. 8,000 tokens free, up to 128,000 for Plus and Pro subscribers. Temporary shutdown of GPT-4o, then return only for paying users. Every decision — part of a grand savings strategy.

Competitive pressure intensifies. Google has stable profit, own data centers and TPUs. Microsoft helps, but it’s not enough. OpenAI is forced to constantly seek financing for training and inference. Under these conditions, efficiency becomes a survival issue. So they survive as they can.

Perhaps the era of smart optimization begins, where engineering elegance matters more than brute force. And GPT-5 is like a manifesto of the new approach.