OpenAI Releases Flagship AI Model GPT-5
GPT-5 is here.
— OpenAI (@OpenAI) August 7, 2025
Rolling out to everyone starting today. https://t.co/rOcZ8J2btI pic.twitter.com/dk6zLTe04s
GPT-5 is the first “unified” neural network that combines sequential reasoning and GPT-style quick answers. A special router determines which approach to take to solve a problem: give a quick answer or 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 now capable of performing tasks on behalf of the user, such as creating software applications, navigating the calendar, or creating research reports.
The startup's CEO Sam Altman called GPT-5 "the best model in the world" and a "significant step" toward creating artificial general intelligence that can outperform humans at the most economically valuable work.
GPT-5 can be used by users without a paid subscription, with certain limits. For Plus and Pro users, these are increased.
ByAPIThree models are available: GPT-5, GPT-5 mini, GPT-5 nano.
OpenAI positions GPT-5 as the most advanced in several areas. It is ahead of Anthropic, Google DeepMind, and xAI in some areas, but behind its competitors in others.
Among the new model’s strengths is programming. In the SWE-bench Verified test, it scored 74.9% on the first try, beating 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 like the Bernoulli effect. It generated hundreds of lines of code in a couple of minutes.
demo time:
— Sam Altman (@sama) August 7, 2025
GPT-5 can make something interactive to explain complex concepts like the bernoulli effect to you, churning out hundreds of lines of code in a couple of minutes. pic.twitter.com/cIU7O608TT
In another, the model created a web app for learning French.
On Humanity's Last Exam, which assesses AI performance in math, humanities, and science, GPT-5 with advanced reasoning (GPT-5 Pro) scored 42%. Grok 4 Heavy scored higher at 44.4%.
Elon Musk took the opportunity to troll OpenAI.
Bottom line though:
— Elon Musk (@elonmusk) August 7, 2025
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/BrggsEwnuz
“Grok 4 Heavy was smarter two weeks ago than GPT5 is now, and G4H is already much better,” the billionaire wrote.
On the GPQA Diamond test, which consists of PhD-level scientific questions, GPT-5 pro scored 89.4% on the first attempt, outperforming Claude Opus 4.1 (80.9%) and Grok 4 Heavy (88.9%).
OpenAI claims that GPT-5 is better at handling health-related questions. In HealthBench Hard Hallucinations, which measures a model’s accuracy on health topics, GPT-5 hallucinated 1.6% of the time. This is much lower than previous models GPT-4o and o3 — 12.9% and 15.8%, respectively.
The company says GPT-5 outperforms other tools in harder-to-measure subjective areas like creative design and writing.
The new model hallucinates much less overall, at 4.8% of the time. This is significantly lower than o3 and GPT-4o, which “invent” false information in 22% and 20.6% of responses, respectively.
On Tau-bench, which measures an AI’s ability to perform simulated online tasks, GPT-5 had mixed results. On the part of the test where you navigate an airline website, the model scored 63.5%. o3 scored 64.8%. On the part where you navigate retail pages, it scored 81.1%, which is lower than Claude Opus 4.1’s 82.4%.
OpenAI noted that the new neural network is more secure: it produces false answers less often and is more effective at identifying intruders.
With the release of GPT-5, ChatGPT now includes a customization feature that allows users to customize the chatbot’s communication style. Users can choose from a range of personality types: Cynic, Robot, Listener, and Nerd. These settings automatically influence the wording of responses, eliminating the need to manually set the desired tone each time.
Among other updates:
- improved voice mode - it has become more natural and intelligent;
- the ability to customize the color of chats;
- Connecting third-party services like Gmail and Google Calendar to get better responses.
As a reminder, in August, OpenAI released open-source reasoning AI models that have shown strong performance in a number of benchmarks and are available for download on Hugging Face.