OpenAI’s GPT-5 is here | TechCrunch

OpenAI has launched GPT-5, a new flagship AI model that will provide electricity to the next generation of chatgpt from the company.

GPT-5, which was released on Thursday, is the first “United” AI model of OpenAi and combines the reasoning options of his O-series models with the rapid reactions from his GPT series. The following generation model signals a new era for chatgpt-and the maker, openly on OpenAi’s wider ambitions to develop AI systems that look more like agents than on chatbots.

Although GPT-4 AI-Chatbots enabled to offer smart answers to a wide range of questions, GPT-5 suggests that ChatGPT performs a wide range of tasks on behalf of user-such as generating software applications, navigating a calendar of a user or making research notes.

With GPT-5, OpenAi has also tried chatgpt easier to use. Instead of asking users to choose the right settings, GPT-5 is equipped with a real-time router who decides how to offer the best answer, whether it responds to questions from users or takes extra time to ‘think’ through answers.

During a briefing with reporters, OpenAi CEO Sam Altman claimed that GPT-5 is ‘the best model in the world’, and said that it was a ‘important step’ represented on the path of the company to develop AI that can perform most for economically valuable work of people-that means artificial general intelligence (AGI).

“Having something like GPT-5 would be almost unimaginable on every last time in history,” said Altman.

From Thursday, GPT-5 will be available for all free Chatgpt users as their standard model. OpenAi’s VP from Chatgpt, Nick Turley, said that this is part of the company’s efforts to give free users access to an AI reasoning model for the first time. (The company previously conquered these more advanced models behind a payment wall.)

“This is just one of the ways I am enthusiastic about living the mission to ensure that this stuff really benefits people,” Turley said about the decision, referring to OpenAi’s long -term mission To spread advanced AI to as many people as possible.

Expectations are high for GPT-5, one of the most expected product launches from OpenAi since Chatgpt placed the company on the map in 2022. Since then, Chatgpt has grown into one of the world’s most popular consumer products, which reach more than 700 million users every week.

Many see GPT-5 as a Bellwether for AI prolarting broadly, and the reception of the model by Silicon Valley could have in-depth consequences for Big Tech, Wall Street and policymakers who regulate technology. These stakeholders look to see if GPT-5 offers a considerable leap in the possibilities of AI, just like its predecessor, GPT-4, who disputed the expectations of what software can do.

GPT-5 offers a slight lead over the competition

OpenAI claims that GPT-5 is state-of-the-art in various domains, which rolls out the leading AI models of Anthropic, Google Deepmind and Elon Musk’s Xai on important benchmarks by Elon Musk. However, GPT-5 performs somewhat deficit at border AI models in other areas.

The company says that GPT-5 offers the border level performance on coding; Altman said that the model excels specifically in spinning entire software applications on request, in what is known as ‘atmospheric coding’.

On SWE-Bank-verified-a test of Real-World coding tasks that are drawn from Github-scoort GPT-5 74.9% on its first attempt. That means that GPT-5 simply performs better than the latest Claude Opus 4.1 model from Anthropic scored that 74.5%and Google DeepMind’s Gemini 2.5 Pro, which scored 59.6%.

On the last exam of humanity and a difficult test that measures AI model performance about mathematics, humanities and the natural sciences scoring a version of GPT-5 with extensive reasoning (GPT-5 Pro) 42% when using tools. That is slightly less than Xai was able to reach with Grok 4 Heavy, who scored 44.4% on the test.

On GPQA Diamond-one test of science questions at PhD-level-GPT-5 Pro, 89.4%scored the first attempt, better than Claude Opus 4.1, which scored 80.9%, and GROK 4 Heavy, who scored 88.9%.

OpenAi says that GPT-5 is better for answering health-related questions. On a test accuracy in AI model reactions on health care topics, Healthbench Hard Hallucinations, says OpenAI GPT-5 (with thinking) Hallucinates only 1.6% of the time. This is much lower than the previous GPT-4O and O3 models of the company, which scored 12.9% and 15.8 respectively.

While AI -Chatbots are not medical professionals, millions of people use them for health advice. In response to this phenomenon, the company says that GPT-5 is more proactive about marking potential health problems and helping users to pars medical results.

Moreover, OpenAI says that GPT-5 is better than other AI models about more difficult to measure, subjective domains, such as creative design and writing. Turley said that GPT-5 responds more naturally and shows ‘better taste’ than other AI models for creative tasks.

“The vibes of this model are really good,” Turley said.

GPT-5 is also more accurate than the earlier models of OpenAi, and the company says that it suffers much less from Hallucinations-the tendency of AI models to come up with information with the O-series models. Hallucinations seemed to get worse in the newest AI Reason models from OpenAi, such as O3, and OpenAi said earlier it didn’t fully understand why it happened.

In reactions to chatgpt prompts, OpenAI discovered that GPT-5 (with thinking) hallucinates and responds with incorrect information 4.8% of the time. That is a significant reduction in O3 and GPT-4O, which score hallucination rates of 22% and 20.6% on the test respectively.

On a benchmark that measures the agentic possibility of an AI model to complete simulated online tasks, Tau-Bench, GPT-5, offers mixed performance. During a part of the test that measures the ability of an AI to navigate the website of an airline, GPT-5 63.5%scores somewhat under performance, which scored 64.8%. On another part of the test that Meases AI’s power to navigate in retail websites, GPT-5 scores 81.1%, underperforming Claude Opus 4.1, which scored 82.4%.

OpenAi also says that GPT-5 is safer than its earlier models. While AI reasoning models occasionally show or lie a tendency to schedule to promote their own goals, OpenAI discovered that GPT-5 was deceptive than other models with a lower pace.

Alex Beutel, the lead of Safety Research of OpenAi, said that reducing deception not only improves the safety of GPT-5, but also the user experience, which creates a model that “is” more transparent and fair in ways in which users can trust. “

Beutel also notes that GPT-5 is better at distinguishing bad actors who try to abuse chatgpt and users who submit harmless requests. This results in GPT-5 being able to refuse more unsafe questions, while offering fewer rejections to users looking for harmless information.

Upgrades for consumers and developers

Chatgpt receives a few user experience upgrades as part of the GPT-5 launch. Users can now choose from four new personalities in the chatgpt setting: cynic, robot, listener and nerd. The company says that they will adjust Chatgpt’s answers without users asking the model specifically to respond in a certain way.

Subscribers to Chatgpt’s $ 20 per month Plus plan receive higher usage limits for GPT-5 than free users. In the meantime, $ 200 a month pro-subscribers have unlimited access to GPT-5, as well as a performed version called GPT-5 Pro that uses extra computational sources to produce better answers. Organizations in the OpenAI, EDU and Enterprise plans team will have access to GPT-5 next week as their standard model.

For developers, GPT-5 comes to OpenAi’s API in three sizes-GPT-5, GPT-5-Mini and GPT-5-nano-Die more or less “reasoning” by tasks. Developers can now also control the Verbinity in the OpenAI API, and decide how long or short the answers of an AI model should be.

The basic model of GPT -5 costs developers $ 1.25 per million input tokens (around 750,000 words, longer than the entire “Lord of the Rings” series) and $ 10 per million output tokens.

The launch of GPT-5 comes after a busy week before OpenAi. The company has released an open-weight reasoning model, GPT-Oss, that developers and companies can download for free and can be carried out with a fraction of the costs. The open model almost corresponded to the possibilities of the earlier top models of OpenAi, O3 and O4-Mini, but GPT-5 sets a new standard for border performance in some areas, such as coding.

However, GPT-5 seems to be roughly on the same footing with other frontier AI models in different areas. Benchmarks of course only tell a part of the story for every AI model, and it is still to see how developers will use GPT-5 in the real world, and whether the model is really a step above the competition.