GPT-5: OpenAI's next generation of artificial intelligence – advances in reasoning and performance – Image: Xpert.Digital
NEXT-LEVEL-KI: This awaits us with GPT-5 in summer
GPT-5 announced: Progress in thinking and performance
The upcoming publication of GPT-5 marks an important milestone in the development of artificial intelligence. According to the latest information, Openai has adapted its publication strategy, whereby GPT-5 is now planned for summer 2025. The recent announcements indicate substantial improvements in thinking, multimodal processing and user -friendliness that go beyond previous models.
Suitable for:
Current level of development and schedule
Openai released GPT-4.5 (code name “Orion”) on February 27, 2025, which serves as a transition model before the introduction of GPT-5. GPT-5 was originally announced for the end of May 2025, but Openai CEO Sam Altman announced a change in these plans on April 4, 2025 .
The new strategy provides to first publish the Reasoning models O3 and O4-Mini in the coming weeks before GPT-5 will follow “in a few months”. There are several reasons for this delay:
"The integration of all components into a uniform system has proven to be more difficult than originally assumed. At the same time, the additional development period enables GPT-5 to make significantly better than we initially thought," said Altman. In addition, Openai takes time to ensure sufficient computing capacities for the “unprecedented demand”.
From GPT-4.5 to GPT-5
GPT-4.5 represents the last model without integrated chain-of-thought technology. In contrast, GPT-5 will fully integrate these thinking skills. Sam Altman described the development as a “standardization of the O series and GPT series into a single, more powerful system”.
Technical progress and performance increases
The technical specifications of GPT-5 indicate an impressive leap in performance. Based on industry information, the following improvements are to be expected:
Model size and computing power
According to current estimates, model size and computing power are experiencing a massive expansion of model parameters. GPT-4 has approximately 1 trillion parameters (1,000B), while GPT-5 is expected to increase to 5 to 10 trillion parameters (5,000B – 10,000B). The dataset size will grow from approximately 10 trillion tokens for GPT-4 to an estimated 40 to 100 trillion tokens for GPT-5. This development will require a significant increase in training hardware, from 25,000 A100 GPUs for GPT-4 to 250,000 to 500,000 H100 GPUs for GPT-5, increasing training costs from approximately $100 million to $1.25 billion to $2.5 billion. This 10- to 20-fold increase in computing power enables GPT-5 to operate significantly faster and more precisely, processing longer and more complex queries with better logic and fewer errors.
Suitable for:
New functions and skills
GPT-5 will not only be more powerful, but also more functional. The most important innovations include:
System-2 thinking and improved logic skills
Unlike previous models, which were primarily based on System 1 (quick and intuitive) thinking, GPT-5 will increasingly implement System 2 thinking – a slower but more rational way of thinking that breaks complex problems down into smaller steps. This enhanced reasoning ability is supported by OpenAI's internal "Strawberry" technology, which enhances mathematical, logical, and problem-solving skills.
Comprehensive multimodal skills
GPT-5 will offer seamless integration of text, language, images and video. This extended multimodal processing enables:
- Language-to-language functionality for more natural conversations
- Improved image processing and generation
- Support for video content
- Coherent processing of different data types
When Altman mentioned in an interview with Bill Gates, the integration of Audio "had a much greater response than expected" and in GPT-5 "this can be promoted much further".
Simplified user experience
A main goal in the development of GPT-5 is the simplification of the user interface. Altman confirmed that the model will “know when it should think for a long time or not, and generally be useful for a very wide range of tasks”.
There is no need to switch between different models. Altman commented: "We hate the model voter too". Instead, GPT-5 will automatically choose the best approach, be it for quick answers or profound analytical considerations.
Personalization and adaptability
GPT-5 will offer significantly better adaptability. Users can adapt the system to their preferences, communication styles and external data sources such as email or calendar.
Reduced hallucinations and improved reliability
Another key benefit is the expected reduction in "hallucinations" – in which AI chatbots fabricate information. The improved logic and reasoning processes should help the model better assess the plausibility of its answers.
Suitable for:
Summer 2025: What GPT-5 promises for users
GPT-5 represents an important step towards Openai's goal of general artificial intelligence. The integration of advanced thinking skills, multimodal functions and improved usability could make AI more accessible to a wider user group.
Although GPT-5 is not yet a real AGI (Artificial General Intelligence), it could fundamentally change the way in which companies and knowledge workers use AI in everyday life. With the expected improvements, both experts and everyday users of AI could trust more.
While we are waiting for the publication in the summer of 2025, the gradual introduction of the O3 and O4 mini models will already give a foretaste of the skills that are fully integrated into GPT-5 in the coming weeks.
Suitable for:
Your global marketing and business development partner
☑️ Our business language is English or German
☑️ NEW: Correspondence in your national language!
Konrad Wolfenstein
I would be happy to serve you and my team as a personal advisor.
You can contact me by filling out the contact form or simply call me on +49 89 89 674 804 (Munich) . My email address is: wolfenstein ∂ xpert.digital
I'm looking forward to our joint project.
