OpenAI has announced the deployment of GPT-4.5. An expected launch, but only by way of intermediate update, which must have continuation “In a few months” With the new GPT-5 generation that should be a greater level jump for these technologies.
GPT-4.5 is the new version of the deep learning model (LLM) that feeds applications such as the artificial intelligence assistant, ChatgPT, and others such as Microsoft co -pilot. Although it is not considered a “avant -garde” model, according to the company itself, it offers significant improvements in computational efficiency, writing quality, knowledge of the world and reduced hallucinations rates compared to its predecessor.
The model is making available as a «Preview of research»which means that it will first undergo a controlled implementation initially focused on collecting comments from users and giving the final touches before it is available for the general public.
They say from Openai that GPT-4.5 has an improvement of more than ten times in computational efficiency compared to GPT-4. This means that the model can achieve similar or better results with significantly less processing power. In this regard and in a related ad, Altman complained that “They had run out of gpus”. It is not known if this is affecting the development (delayed) of GPT-5 and the influence of the emergence of the Chinese Deepseek model and its promises to achieve a more efficient, more open, more accessible and much more economical.
One of the main statements is that GPT-4.5 has improved the “EQ” or emotional intelligencewhich makes it more similar to a human in their answers. Openai says that nuances understands better, which could make it a stronger tool to write, code and solve problems. This should be extremely valuable for all users.
Another key objective of development has been to minimize ‘hallucinations’, cases in which the model safely generates false or meaningless information. For this, the model has trained using a combination of «New supervision techniques» together with established methods such as the supervised fine adjustment (SFT) and reinforcement learning from human feedback (RLHF).
Despite these advances, Openai is clear that GPT-4.5 is not an avant-garde model and, for some things, it may not even be a great step forward in any way. It does not present any new or significantly improved capacity And, in fact, you can really have a performance “Below of O1, O3-Mini and deep research in most preparation evaluations”. Everything must change with GPT-5, in accelerated development, but without a release date.
GPT-4.5 is only available at the time for Chatgpt Pro members who pay 200 dollars per month. Next week will be available for chatgpt plus ($ 20 per month) subscribers. It is not clear if this model will be implemented for free users or on what date.
