Openai’s GPT-5 here | Techcrunch

Openai GPT-5, a new model of artificial intelligence that will operate the next generation of Chatgpt for the company.

GPT-5, which was released on Thursday, is the first “unified” model of artificial intelligence in Openai and combines thinking capabilities The chains O. From models with fast responses to them GPT series. The next generation model refers to a new era for ChatGPT-creator, Openai-indicates the broader Openai’s aspirations to develop artificial intelligence systems Like the agents From Chatbots.

while GPT-4 AI Chatbots empowerment to provide smart responses to a wide range of questions, GPT-5 Chatgpt allows completing a wide range of tasks on behalf of users-such as generating software applications, navigating user calendar, or creating research summaries.

With GPT-5, Openai also sought to make Chatgpt more simple to use. Instead of asking users to choose the correct settings, GPT-5 comes with a actual time orientation device that decides how to provide the best answer, whether it responds to the user’s questions quickly or spending additional time “thinking” through answers.

During a briefing with the correspondents, the CEO of Openai Sam Altman claimed that the GPT-5 is the “best model in the world”, and he said it represents an “important step” on the company’s path to develop artificial intelligence that can outperform humans in most work of economic value-iGI.

“The presence of something like GPT-5 will be largely unreasonable in the history,” Altman said.

Starting on Thursday, GPT-5 will be available to all free users for Chatgpt as a virtual model. This is part of the company’s efforts to give free users access to the artificial intelligence thinking model for the first time. (Previously, the company has placed these most advanced models behind Paywall.)

“This is just one of the ways I am excited to live in the mission, and make sure that these things are already benefiting people,” Torley said in the decision. a task To distribute advanced artificial intelligence to the largest possible number of people.

Expectations are high for GPT-5, which is one of the most expected products in Openai since Chatgpt has placed the company on the map in 2022. Since then, ChatGPT has grown to become one of the most popular consumer products in the world, reaching more than 700 million users every week-nearly 10 % of Globe population, according to the company.

Many see that GPT-5 as a widespread action bell, and the form of the form by the Silicon Valley can be deep effects on major technology, Wall Street, and policymakers who organize technology. These stakeholders are monitoring these to see if GPT-5 offers a great leap in the capabilities of artificial intelligence, such as its predecessor, GPT-4, which challenges the expectations of what the programs can do.

GPT-5 provides a slight feature of competition

Openai claims that the GPT-5 is in the latest model in many areas, as it leads a little bit of artificial intelligence models of anthropor, Google Dibdend and Xai Xai on the main standards. However, the GPT-5 is slightly vulnerable to the border AI models in other regions.

The company says that GPT-5 performs the border levels around the coding level; Al -Taman said that the model is particularly superior to rotating the entire program applications upon request, in what is known as “vibrant coding”.

On Swe-Bused- Test of the real world coding tasks has been withdrawn from GitHub-GPT-5 degrees 74.9 % in its first attempt. This means that the GPT-5 surpasses the latest Claude Obus 4.1 from the Anthrop. 74.5 %And Google DeepMind’s Gemini 2.5 Pro, which recorded 59.6 %.

In the last humanity exam-a difficult test, measures the performance of the artificial intelligence model through mathematics, humanities and natural sciences-a copy of GPT-5 was recorded with 42 % GPT-5 Pro when using tools. This is just less than Xai that you managed to achieve with Groc 4 heavyWhich recorded 44.4 % in the test.

On GPQA Diamond- Science Question Test at PhD-GPT-5 Pro 89.4 % in his first attempt, beats Claude Obus 4.1, which recorded 80.9 %, and Grok 4 Heavy, which recorded 88.9 %.

Openai says GPT-5 is better to answer health questions. In a test to measure accuracy in the artificial intelligence model responses on health care topics, Openai Healthbench says, GPT-5 (with thinking) only 1.6 % of time. This is much lower than the former GPT-4O and O3 models of the company, which recorded 12.9 % and 15.8, respectively.

While Ai Chatbots are not medical professionals, Millions of people They use it for health advice. In response to this phenomenon, the company says that GPT-5 is more active in setting a sign of potential health concerns and helping users analyze medical results.

In addition, Openai says that GPT-5 is better than other artificial intelligence models in more difficult to check, Self -fields, Like creative design and writing. Turley said that GPT-5 responds naturally and offers a “better taste” than other artificial intelligence models in creative tasks.

“The reactions of that model are really good,” said Torley.

GPT-5 is also more accurate than the previous models of Openai, and the company says it suffers from hallucinations much lower-tendency to artificial intelligence models to replace information-compared to the Series Models O. It seems hallucinations It gets worse In the latest thinking models of artificial intelligence in Openai, such as O3, and Openai previously, she said she had not fully understood the reason for her occurrence.

In the responses to ChatGPT claims, Openai found that GPT-5 (with hallucinogenic thinking) and responds with incorrect information 4.8 % of time. This is a significant decrease from O3 and GPT-4O, which records hallucinations 22 % and 20.6 %, respectively, in the test.

On the standard of measuring AI’s agent’s ability to complete online simulation tasks, GPT-5, GPT-5 offers mixed performance. On a portion of the test that measures the ability of artificial intelligence to navigate at the airline website, GPT-5 achieves 63.5 %, which is slightly weak O3, which recorded 64.8 %. On another part of the test that measures the ability of artificial intelligence to navigate at retail sites, GPT-5 is 81.1 %, as Claude OPUS 4.1, which recorded 82.4 %.

Openai also says that GPT-5 is safer than its previous models. While the thinking models of artificial intelligence sometimes show a tendency to Human planned Or lie to enhance their own goals, Openai found that GPT-5 was deceptive at a rate less than other models.

Alex Biotel, the Openai Safety Research Command, said that reducing deception not only improves the safety of GPT-5, but also the user experience, which creates a more transparent and honest model in ways that users can trust.

Beutel also notes that GPT-5 is better to distinguish between bad actors trying to misuse Chatgpt and users who make harmless requests. This is the result of GPT-5 capable of rejecting more unsafe questions, with fewer rejection operations to users looking for harmless information.

Promises for consumers and developers

Chatgpt gets some user experience promotions as part of the GPT-5 launch. Users can now choose from four new characters in the preparation of Chatgpt: satirical, robot, listener, and NERD. The company says that this will adapt to the Chatgpt responses without asking users to demand specifically the form to respond in a certain way.

Participants in the ChatGPT plan, which cost $ 20 per month, gets higher use of the GPT-5 free users. Meanwhile, subscribers in $ 200 in two months will have unlimited arrival to GPT-5, as well as a wavy version called GPT-5 Pro uses additional calculation resources to produce better answers. Organizations in the Openai, EDU and Enterprise team will be able to reach GPT-5 as a virtual model next week.

For developers, GPT-5 comes to API in Openai in three sizes-GPT-5, GPT-5-MINI and GPT-5-nano-which will spend time or less “thinking” through tasks. Developers can also control API Openai, which determines the duration in which the responses of the artificial intelligence model or palaces must be.

GPT-5 basic model will cost $ 1.25 per million input codes (approximately 750,000 words, longer than the entire “Lord of the Rings” series) and $ 10 per million directing symbols.

The GPT-5 launch comes after a crowd for Openai. The company released an open -weight thinking model, GPT -SSDevelopers and institutions can download it for free and operate it in a small part of the cost. The open model was almost identical to the first previous Openai capabilities, O3 and O4-MINI, but GPT-5 sets a new standard for border performance in some areas, such as coding.

However, the GPT-5 appears to be equal with other border AI models in several regions. The criteria, of course, only tell part of the story of any Amnesty International model, and it remains to see how the GPT-5 developers will use in the real world, and whether the model is really a step above the competition.