Openai released a flagship shi model GPT-5

Openai has launched a new flagship shi model that will be the basis of the next generation of Chatgpt.

GPT-5 IS HERE.

Rolling Out to Everyone Starting Today.https: //t.co/rocz8j2bti pic.twitter.com/dk6zlte04s

– Openai (@openai) August 7, 2025

GPT-5 is the first “unified” neural network that combines sequential reasoning and quick GPT-style. The special router determines which approach to apply for the task: to give a quick response or to spend more time thinking to improve the quality of the result.

GPT-4 allowed the chatbot to answer a wide range of questions. GPT-5 is already capable of performing tasks on behalf of the user, including creating software applications, working with a calendar or forming research reports.

GPT-5 input and output tokens. Source: X.

Benchmarks Openai positions GPT-5 as an advanced in several directions. By a number of indicators, it is ahead of the development of anthropic, Google Deepmind and XAI, although in some areas it is inferior to competitors.

Among the strengths of the new model are programming. In the Swe-Bench Verified test, it scored 74.9%on the first attempt by bypassing Claude Opus 4.1 (74.5%) and Gemini 2.5 Pro (59.6%).

In one example, GPT-5 created interactive material to explain complex concepts, including the Bernoulli effect, generating hundreds of code lines in a few minutes.

Demo Time:

GPT-5 can Make Something Interactive to Explain Concepts Like The Bernoulli Effect To You, Churning Out HundDS of Lines of Code in A Couple of Minutes. pic.twitter.com/ciu7O608TT

– Sam Altman (@sama) August 7, 2025

Otherwise, the model has created a web -makeup for learning French.

In the Humanity’s Last Exam test, which evaluates the work of AI in mathematics, humanities and natural sciences, GPT-5 Pro scored 42%. Grok 4 Heavy showed a higher result – 44.4%.

Source: Openai.

Ilon Musk did not refrain from Openai’s trolling.

Bottom Line Thought:

Grok 4 Heavy Was Smarter 2 weeks ago than gpt5 is now and g4h is already a Lot Better.

Let that sink in. https://t.co/bgggsewnuz

– Elon Musk (@Lonmusk) August 7, 2025

“Grok 4 Heavy was smarter two weeks ago than GPT-5 now, and G4H is much better,” the billionaire wrote.

In the GPQA Diamond test, which contains doctoral scientific issues, GPT-5 Pro scored 89.4%on the first attempt by exceeding Claude Opus 4.1 (80.9%) and Grok 4 Heavy (88.9%).

Openai claims that GPT-5 works better with health topics. In Healthbench Hard Hallucinations, which measures the accuracy of medicine answers, GPT-5 invents data in 1.6% of cases. This is much less than GPT-4O (12.9%) and O3 (15.8%).

In general, the new model “hallucin” in only 4.8% of answers, which is much lower than O3 and GPT-4O-22% and 20.6%, respectively.

In Tau-Bench, which checks the ability of AI to perform simulated online tasks, GPT-5 showed mixed results: 63.5% in the scenario with the airline (against 64.8% in O3) and 81.1% in the task with retailers (against 82.4% in CLAUDE OPUS 4.1).

Openai emphasizes the increased safety of GPT-5: it less often gives out false answers and better identifies malefactors.

Renewal Along with the GPT-5 release, Chatgpt has a customization feature that allows you to set up a chatbot communication style. Users can choose one of the types of personality: cynic, robot, listener or botan. These parameters automatically affect the formulation of answers, eliminating the need to set the tone every time manually.

Among other innovations:

• improved voice regime – more natural and “smart”;
• possibility of changing the color of chats;
• Integration with third -party services, including Gmail and Google Calendar, for more relevant answers.

Recall that in August Openai released an open code, which show high results in a number of benchmarks and are available for download on Hugging Face.

The gun