GPT-5 is officially available and from today it is possible to use the new model based on the artificial intelligence of Openai directly on Chatgpt. Openai defines him his most advanced model ever and made it available not only to subscribers, but also to free users, despite some limitations of use. Compared to previous versions, it offers more precise and faster answers.
Sam Altman, CEO of Openai, compares him to an epochal leap compared to previous models. According to him, «With GPT-3 it seemed to speak with a high school student (…), with GPT-4 it seemed to speak with a university student (…), but with GPT-5 it is the first time that he really seems to speak with a PhD level expert». An interesting feature of GPT-5 is that the model is proposed within chatgpt in a single integrated version, capable of automatically switching to an advanced reasoning mode when necessary, without the user having to select it manually. Next week, among other things, Openii will also present the Enterprise and Ed versions.
Those who use the free version, once a certain (not well specified) threshold of requests will pass, will move on to GPT-5 Mini, faster but less powerful. In addition to linguistic and problem solving skills, they stand out customization functions with four selectable “personality” (cynic, robot, listener and nerd) and the possibility of changing the color of the chats. In the internal tests, GPT-5 has passed any previous coding benchmark, also allowing the instantaneous creation of interactive websites. On the security front, Openii has carried out over 5,000 hours of assessments to reduce “hallucinations” and deceptive responses, introducing the so -called “Safe CompleTations” (or “safe completions”), that is, useful but safe answers to potentially risky questions.
How it works chatgpt with gpt-5
GPT-5 was designed to offer a more fluid and “human” experience in the interaction, eliminating the need to manually pass between different models. This choice simplifies use and reduces errors due to users wrong by users. The technology of routing Internal decides whether to use the complex reasoning mode according to the application, or whether to maintain a lighter processing for quick responses. This means that even those who are not familiar with AI models can benefit from advanced skills without having to worry about using specific settings or commands.
One of the most interesting news is the practical application in the field of programming. In the tests, GPT-5 managed to generate hundreds of rows of code working in a few seconds, demonstrating a clear improvement compared to previous models in sector benchmark such as Swe-Bench And Aider Polyglot. These performances open the way to what Altman calls “on demand software», That is, the instantaneous creation of personalized applications without the intervention of a human programmer.
As for safety, the model was trained to recognize when a request can have harmful implications, trying to provide useful information but in a format that cannot be exploited for illegal purposes. This ability to modulate the level of detail of the responses, keeping the conversation safe, represents an important step in the ethical management of the AI. Alex Beutel, head of model safety research, explained in this regard:
If someone asks (a chatgpt): ‘How much energy is necessary to set a certain material?’, Could it be an attacker who tries to get around the safety protections and cause damage. Or it could be a student who asks a scientific question to understand the physics of this material. This creates a real challenge to find the best way the model can respond. With the “safe completions”, GPT-5 tries to provide an answer as useful as possible, but within the limits of safety.
In addition, GPT-5 is more inclined to admit when he does not know an answer, thus reducing the risk of disseminating incorrect information.
Who can use chatgpt with gpt-5
As for the availability of GPT-5, access to the new model is already guaranteed to all chatgpt users, with differences only within the limits of use: pro and plus subscribers can use it practically without restrictions, while those who use the free version will have a certain number of complete interactions before automatically moved on to the mini version. Openii did not specify how many interactions can count free users before moving on to GPT-5 mini.
