- OPENAI seems to test GPT-5 on the basis of disclosed files and internal biosecurity tools
- According to rumors, the GPT-5 would unify memory, reasoning, vision and completion of the task in a
- No specific release date has been announced, but it is likely to officially deploy in the coming months
Rumors that OpenAi is already testing the GPT – 5 have started to spread through leaks and advice appearing online. A particularly notable example was shared on X by the engineer Tibor Blaho when he published a partial screenshot of a configuration file referring to “GPT – 5 Alpha resttings”, dated July 13, 2025.
This same week, independent researchers have discovered a mention of GPT-5 in the Biosec internal benchmark for OpenAI, suggesting that the model is already tested in sensitive fields such as biosecurity.
In the event that these indirect omens were not enough, Xikun Zhang of Openai explicitly declared that GPT-5 “arrives” during a discussion on the new feature of Agent Chatgpt.
The attention paid to the release date of GPT-5 is not surprising because people asked Openai on this subject since the company published the GPT-4 in 2023, and only accelerated when GPT-4.5 was released last year. Chatgpt agents have undoubtedly brought speculation to a fever argument, because these digital assistants can go out and do things online to reserve tickets and organize calendars. These are all things that people have thought that GPT-5 would bring to Chatgpt.
GPT-5-SEASONING-ALPHA-2025-07-13H / T @SWISHFEVER PIC.TWITTER.COM/HQ5CSD0IC6July 19, 2025
GPT-5 Future
Comments from people using chatgpt agents can actually be used to complete the GPT-5 training. And it will take a lot of training if rumors on the context window at a million tonnes are true. The same goes for the supposedly unified nature of GPT-5.
This would mean that GPT-5 will not only include the switch between features such as visual analysis, code interpretation and the same capacities as the chatgpt agent; It will work as a singular system. You can ask him to interpret an image, send an email, plan a meeting and compose and make a vocal summary from a single prompt.
For example, a parent could coordinate school schedules, meal plans and the logistics of the last-minute birthday party, or you could plan a trip, book hotels, put it in your calendar and send an email to your family the details of a single request. The GPT-5 is supposed to be designed to resolve the problems of halvery hallucination and misunderstandings, allowing people to trust it more than GPT-4 or GPT-4.5.
There is also the question of memory. Openai discreetly deployed long-term memory in the Chatppt, which means that the model remembers things about you, but the GPT-5 could make it even more powerful.
Of course, security issues are already boiling. The mention of GPT-5 tested in biosecurity contexts has frightened people. If he can reason well enough about biology to help complex research, could she also spit dangerous information if she was invited to the wrong sense? OPENAI has promised to build guarantees, but history tells us that people are very good at finding smart ways around digital fences.
Anyway, we will all know it early enough, although on the basis of previous launches, it will probably be limited to subscribers of the higher chatgpt levels at the start. But you can bet Openai will not publish GPT-5 as long as he is sure that he will not embarrass them on the first day.
