Introduction
The development of language models reached a new milestone with the arrival of GPT-5, launched by OpenAI in August 2025. Since then, the pace has not slowed: by the end of 2026, OpenAI, Anthropic and Google had each released a new generation of models. In this article, we look back at the evolution of large language models (LLMs), their business applications and what this race concretely changes for you. To keep up with news on AI applied to business, find all our articles on the diagonall website.
The useful question for an executive is not "which model is the best this month?", but "how do we build tools that benefit from each new model without starting over?". That is what we do with our AI systems by business function.
The Evolution of Language Models
Language models have evolved considerably since GPT-3. GPT-4 brought clear improvements in understanding context and in the coherence of texts. GPT-5 then made "reasoning" mainstream: the model spends more time on difficult questions and answers simple ones faster.
What GPT-5 Changed
GPT-5 stood out for its ability to follow complex instructions, use tools and chain together several steps of work. This is what made AI agents possible in business: programs that no longer just answer, but read a document, query a piece of software and prepare an action.
Where Do Things Stand at the End of 2026?
GPT-5 is no longer OpenAI's latest model: GPT-6 was launched in September 2026. Anthropic has released the Claude 5 generation starting in summer 2026, and Google unveiled Gemini 4 in late September 2026. Each generation is making progress along the same lines: longer tasks carried out without help, better use of tools and software, fewer errors on professional documents.
For a company, the practical consequence is simple: a model chosen today will be outdated in a few months. It is better to have tools designed to switch models easily than tools tied to a single provider.
Practical Applications
- Customer service automation: chatbots built on these models handle more complex conversations and know when to hand over to a human.
- Optimization of internal processes: by connecting the model to company documents (RAG) or adapting it to a specific task, you get answers based on your own data.
- AI agents: assistants that carry out a series of tasks in your software, under human supervision.
Challenges and Ethical Considerations
With the increased power of LLMs come significant ethical and regulatory challenges. The AI Act already partly applies: the rules on general-purpose AI models since August 2025, and the transparency obligations since August 2026. Companies must also put internal rules in place to govern the use of these tools.
The Future of LLMs
The labs are already working on the next generations. The announced directions remain the same: more reliable reasoning, a better understanding of the context of each business function, and agents able to work longer without supervision.
Towards More Responsible AI
As LLMs evolve, it becomes essential to use them responsibly. Developing usage rules and governance frameworks, as the article on AI governance shows, is a priority for companies and regulators alike. Training teams is part of this: see our AI training courses.
💡 Did you know? diagonall is a French AI agency that supports companies in their AI transformation with training, strategic consulting and tailor-made solutions.
Conclusion
GPT-5 marked a milestone in the evolution of language models, and the next generation is already here. Companies stand to gain by adopting these technologies, provided they put proper safeguards in place and build tools that survive model changes. LLMs will continue to play a central role in the transformation of businesses.
Frequently Asked Questions
What were the main improvements of GPT-5 over GPT-4?
GPT-5 made step-by-step reasoning mainstream and improved the following of complex instructions and the use of tools, which paved the way for AI agents.
Is GPT-5 still the latest model?
No. OpenAI launched GPT-6 in September 2026, and its competitors have also released new generations (Claude 5 at Anthropic, Gemini 4 at Google).
How can companies ensure ethical use of these models?
By setting clear usage rules, training their teams and complying with regulations such as the AI Act and the GDPR.
Ready to take action?



