It has cost, but Openai He just launched GPT-5. And best of all, unlike their predecessors, which at first were only available for subscriptions, it’s free for everyone.
GPT-5 is the first unified AI model of Sam Altman’s company. That is, it combines the capabilities of the models of reasoning O- with the rapid responses and conversation skills of the GPT- models. One more step towards the AGIgeneral artificial intelligence. That is, an AI that can do everything, like humans.
The new large language model GPT-5 It also differs from its predecessor in which It is less and less chatbot and more AI agent. Answering doubts is already a first generation thing. GPT-5 is better to complete complete tasks, from creating an app from scratch to investigate.
GPT-5, an AI that is not the best in everything
In a talk with the American press, Sam Altman has said that GPT-5 is «the best in the world», in global terms. But as you collect TechCrunch, In the benchmarks them Openai has facilitated, there is not always the first.
Of course, he does many things in the background that are not seen, but that are much smarter than in their predecessors. For example, You can choose the language model according to what you ask, or change your personality according to conversation. For the user it is also easier to handle, because you no longer have to choose a specific AI, according to the task.
OpenAI ensures that GPT-5 also Detects better when someone is asking for something with bad intentions. If they ask for instructions that could create an explosive, cause damage, or prepare a scam, even if they are subtle requests, it detects it better than before.
Where he has improved is in the hallucinations. According to data from the GPT-5 company itself, it offers incorrect data 4.8% of the time, compared to 22% of O3 and 20.6% of GPT-4O. He also ensures that «lies» less, in the sense that he falsifies less data to give him the reason, as their predictors sometimes do.
And how do you compare with respect to competition? Benchmarks say that there is no significant leap, and even in some, it loses with Grok, Claude, or previous models.
What is best given is to create code from zero. In the Swe-Bench Verified test, GPT-5 reaches a score of 74.9%, compared to 74.5% of Claude Opus 4.1 of Anthropic or 59.6% of Gemini 2.5 Pro of Google.
However, with the Humanity’s Last Exam test, which measures the performance in science, mathematics and humanities, GPT-5 Pro stays at a 42% score, behind the 44.4% of Grok 4 Heavy. It exceeds everyone in the GPQA Diamond test, with knowledge of doctoral level sciences.
His agricultural skills are not top. With Au-Bench, which measures the ability to navigate for flight websites, it is below O3. Moving on store websites, it does not exceed Claude Opus 4.1.
Where has great improved is in medical and health issuesa section where chatbots are receiving many criticisms. In the Healthbench Hard Hallucinations test, only hallucin in 1.6% of the data, compared to 12.9% of GPT-4O and 15.8% of O3.
Openai also ensures that GPT-5 is superior in aspects that cannot be measured with Bechmarks, such as writing or creative design.
Different versions of GPT-5, free for all
For the first time, Sam Altman’s company offers GPT-5 for free for everyone from dayalthough logically payment users have fewer limitations.
Plus succipers can use many more requests per day than free, and the PRO get unlimited access, and the professional version, GPT-5 Pro.
As Github filtered ahead of time this morning, GPT-5 is available in different versions:
- GPT-5. Manager for the design of logical tasks and, at the general level, which can use users for less specific tasks.
- GPT-5-mini. A light version of the previous model, with a lower reasoning capacity, although ideal for more tight budgets.
- GPT-5-nano. The ideal model for tasks that require higher processing speed, as well as applications that need low latency.
- GPT-5-chat. It has been designed for conversations in the purest style of a chatbot, especially indicated to hold conversations with a natural and multimodal style.
As we see, GPT-5 does not involve a spectacular leap in terms of performancealthough there are improvements in code creation, hallucinations, creativity, and management of health issues.
Its main virtue is its status as a unified AI model, which is worth for everything. That seems to be the future of language models, one more step does the AGI.
Know How we work in NoticiasVE.
Tags: Artificial, viral intelligence, industry