Perhaps due to the disappointing results, Altman had previously written that GPT-4.5 will be the last of the Openai traditional AI models, with GPT-5 planned to be a dynamic combination of LLM of “non-reasoning” and simulated reasoning models such as O3.
A stratospheric price and a technological end end
And about that price, it is a disaster. GPT-4.5 costs $ 75 per million input tokens and $ 150 per million production tokens through the API, compared to the entrance tokens of $ 2.50 per million GPT-4O and tokens out of $ 10 per million. (Tokens are data fragments used by AI models for processing). For developers who use Operai models, this price makes GPT-4.5 practical for many applications where GPT-4O already works properly.
On the contrary, the OpenAI flagship reasoning model, O1 Pro, costs $ 15 per million input tokens and $ 60 per million production tokens, significantly less than GPT-4.5 despite offering specialized capabilities of simulated reasoning. Even more surprising, the O3-mini model costs only $ 1.10 per million input tokens and $ 4.40 per million output tokens, which makes it cheaper than even GPT-4O, while providing much stronger yield in specific tasks.
Operai probably knows about decreased yields in LLMS training for some time. As a result, the company spent most of the last year working on simulated reasoning models such as O1 and O3, which use a different approach to inference time (execution time) to improve performance instead of throwing increasingly large amounts of training data in GPT style models.
While this seems bad news for Openai in the short term, competition is thriving in the AI market. Anthrope’s Claude 3.7 sonnet has demonstrated much better performance than GPT-4.5, with an architecture, according to reports, a more efficient architecture. It is worth noting that the sonnet Claude 3.7 is probably a system of AI models that work together behind the scene, although Anthrope has not provided details about its architecture.
For now, it seems that GPT-4.5 can be the last of its kind, a technological dead end for a non-supervised learning approach that has raided the way for new architectures in AI models, such as the reasoning of the inference time of O3 and perhaps even somewhat more novel, as diffusion-based models. Only time will say how things end.
GPT-4.5 It is now available for Chatgpt Pro subscribers, with PLUS deployment and team subscribers planned for next week, followed by Enterprise and Education customers the following week. Developers can access it through the various OpenAI API at payments, although the company is not sure of its long -term availability.
#lemon #Openais #largest #model #reaches #mixed #revisions
