Artificial intelligence has just crossed a new course, and this time, it is a storm that promises to be. GPT-4.5, the latest creation of Openai, crushes the tracks with a disproportionate ambition: dominate by size. While 2025 was already rose up under the announcements of rival models – such as the Claude d'Anthropic or the DEEPSEEK R1, designed with a derisory budget -, OpenAi put everything on gigantism. With a chatgpt pro subscription at 200 monthly dollars, initiates can now test this colossus. But behind the promises hide daring strategic choices … and surprising paradoxes.
The excess as a doctrine for AI?
Openai no longer hides his ambitions: “bigger, always bigger”. The launch of GPT-4.5 is part of a logic of technological escalation, where each parameter, each artificial neuron, is a step towards an alleged ia ia. Unlike Deepseek, which shaken Silicon Valley by optimizing costs and performance, Openai persists: invested billions, pharaonic infrastructure, and a model so voracious in calculation that it becomes mysterious. “The figures? They will stay in the shadows, ”we slip in OPENAI.
However, this excess would have a merit: capturing the elusive shades of the human. Mia Glaese, head of the alignment team, assures him: “The more things we know, the less we invent. Translation: GPT-4.5 would hallucinate less, because his mass of knowledge prevents him from filling the whites with imagination. An attractive argument, but which raises a question: how far can we grow a model before hitting physical or ethical limits?
Between the lines, a conflict takes shape. On the one hand, supporters of efficiency, like Deepseek, prove that performance is not just a matter of size. On the other, Openai camps on his position, making the raw power a standard.
A risky bet, while academic benchmarks reveal weaknesses: surpassed in math by the O3-minini, followed in science, GPT-4.5 really shines only in language. “Benchmarks do not tell everything,” retorts Yoke. The real revolution would be elsewhere.
User experience: between promises and gray areas
Let's move on to concrete things. Pro subscribers discover GPT-4.5 with promising features: web search, file analysis, image processing … but still no vocal AI. Strange, when you know that multimodal is supposed to be the future. Result: the exchanges remain confined to the text, with the hope that this famous “natural” is finally there.
The first feedback? Mixed. For programming or writing, it is more fluid, more effective. But as soon as it is a question of dealing with sharp scientific subjects, the results are perplexed. Clearly, the evolution is subtle, almost imperceptible for the general public.
The real question is the continuation. OPENAI assures him: GPT-4.5 will be the last model without “chained reasoning”. In other words, no need to manually choose a model: Chatgpt will automatically adapt to each request. An attractive advance … but which raises a real question: where did user freedom go in all of this?
With GPT-4.5, Openai persists in the idea that an ever larger model is the key to an “human” AI, despite the exploding costs. But faced with lighter and equally efficient models, this size race makes you think. Meanwhile, tech giants take the opportunity to increase Wall Street to delusional peaks.
Maximize your Cointribne experience with our 'Read to Earn' program! For each article you read, earn points and access exclusive rewards. Sign up now and start accumulating advantages.
