- OpenAI seems to be testing GPT-5 based on filtered files and internal biosafety tools
- It is rumored that GPT-5 unifies the memory, reasoning, vision and completion of the task in one
- A specific launch date has not been announced, but it is likely that officially deployed in the coming months
The rumors that OpenAI is already testing GPT – 5 have begun to spread thanks to the leaks and suggestions that appear online. A particularly remarkable example was shared in X by the engineer Tibor Blaho when he published a partial screenshot of a configuration file that suggests “GPT -5 Reasoning Alpha”, dated July 13, 2025.
That same week, independent researchers discovered a mention of GPT-5 in the OpenAi’s internal BIOSEC reference repository, suggesting that the model is already being tested in sensitive domains such as biosecurity.
In the event that these indirect portents were not enough, Xikun Zhang of OpenAi explicitly said that GPT-5 “comes” during a discussion about the new chatgpt agent function.
The attention paid to the launch date of GPT-5 is not surprising since people have been asking OpenAi about it since the company launched GPT-4 in 2023, and only accelerated when GPT-4.5 left last year. It could be said that Chatgpt agents brought speculations to a high point, since these digital assistants can go out and do things online for you, such as reserving tickets and organizing calendars. Those are all the things that people have thought that GPT-5 would bring to Chatgpt.
GPT-5-razoning-alpha-2025-07-13h/t @Swishfever pic.twitter.com/hq5csd0ic6July 19, 2025
GPT-5 Future
The comments of people who use ChatGPT agents can be used to complete GPT-5 training. And you will need a lot of training if the rumors about the context window of one million token are true. The same goes for supposedly unified nature of GPT-5.
That would mean that GPT-5 will not only include the change between characteristics such as visual analysis, the interpretation of the code and the same skills as the chatgpt agent; It will work as a singular system. You can ask you to interpret an image, send an email, schedule a meeting and compose and make a vocal summary from a single message.
For example, a father could coordinate school schedules, meal plans and the logistics of the last minute birthday party at the same time, or can plan a trip, book hotels, put it on their calendar and send an email to your family the details of a single application. GPT-5 is supposedly designed to address hallucination problems and nuanced misunderstandings, allowing people to trust it more than GPT-4 or GPT-4.5.
There is also the issue of memory. Operai has been silently implementing long-term memory in Chatgpt, which means that the model remembers things about you, but GPT-5 could make it even more powerful.
Of course, security questions are already bubbling. The mention that GPT-5 be proven in biosecurity contexts has scared people. If you can reason about biology well enough to help with a complex investigation, could you also spit dangerous information if requested incorrectly? Operai has promised to build safeguards, but history tells us that people are very good to find intelligent forms around digital fences.
Anyway, we will all discover it soon, although according to the previous releases, it is likely to be limited to the subscribers of the highest levels of Chatgpt at the beginning. But you can bet on OpenAi that will not launch GPT-5 until they are sure that you will not be ashamed of the first day.
You may also like
