The CEO of the company Sam Altman called GPT-5 “a significant step along the path to Agi” during a briefing of Wednesday’s press. While he stopped claiming that the model reaches general artificial intelligence, Altman has noticed that the latest version is “clearly a generally intelligent model”. He added that GPT-5 is still missing the key traits that would make him reach Agi, a significantly loose term that is defined in Openi paper as “a highly autonomous system that exceeds human beings in economically precious work”. For example, the model still lacks the ability to learn continuously after distribution.
Openii says that GPT-5 is smarter, faster, more useful and more accurate, with a lower hallucination rate than previous models. In typically high terms, Altman compared the GPT-4 to GPT-5 jump to the passage of the iPhone from Pixelato to a retina display. “GPT-5 is the first time that he really seems to talk to an expert in any topic, like a doctoral expert,” said Altman.
In chatgpt, most users will no longer have to choose between the models, since the chat interface now automatically instracts to the right version depending on the complexity of the query and the user’s subscription level. As part of the Thursday launch, Openi has announced two new model variants: a slight GPT-5-Mini and a GPT-5-Nano even faster, cheaper (which is only in the API). According to Openi, free users will have access to GPT-5 and GPT-5-Mini, while the Plus subscription includes the same models with “significantly higher” use limits. Openii says that the professional level of $ 200 per month offers unlimited GPT-5 access, together with GPT-5-Pro, a more powerful version of the GPT-5 model and thought, which allows the model to process a query longer than usual. Professional users will still have access to choose Legacy models.
Starting from next week, Pro users will be able to connect their Gmail, Google contacts and Google calendar to chatgpt, with other levels that obtain access to an unspecified date. “Chatgpt knows automatically when it is more relevant to refer to them, so it is not necessary to select them before chatting,” said the company in one and -mail.
Users can also choose a chat color and select between four preset-cyclists, robot, listener and nerd-un behavior of the Wired newsletter model reported last week. According to the announcement of the Openai blog, it provides to cook these personalities in advanced vocal mode.
The company’s bees will offer users all three models, together with optional checks to activate detailed or direct responses. GPT-5 can keep more information than previous models: it has a context window of 256,000 token, an increase in the context window of 200,000 token available in the previous O3 model of the company. This means that it can better understand long conversations, documents or the code without losing the trace of the context.
Openai says that GPT-5 beats its previous models on different coding benchmarks, including Swe-Bench (scoring 74.9 percent), Swe-Lancer (Phinking GPT-5 marked 55 percent) and Aider Polyglot (he obtained a score from 88 percent), who test the ability of the model to repair the bugs, complete the activities of the activities of Coding quickly and work through multiple programming languages.