- Openai seems to test GPT-5 based on leaked files and internal biosecurity tools
- GPT-5 is joined to unite memory, reasoning, vision and task ending in a
- No one has announced any specific release date but it will probably be officially rolled out in the next few months
Rumors that Openai is already testing GPT -5 has begun to spread thanks to leaks and tips that appear online. A particularly remarkable example was shared on X by Engineer Tibor Blaho when he published a partial screenshot of a configuration file that suggested “GPT -5 -Reasoning Alpha”, dated July 13, 2025.
The same week, independent researchers discovered a mention of GPT-5 in Openai’s internal biosec-benchmark archive, which suggested that the model has already been tried on sensitive domains as biosecurity.
In the event that these indirect portents were not enough, Openai’s Xikun Zhang explicitly said that GPT-5 “comes” coming “during a discussion of the new chatgpt agent function.
The attention directed at GPT-5’s release date is not surprising when people have asked Openai about it since the company released GPT-4 in 2023, and only accelerated when GPT-4.5 came out last year. The Chatgpt agents undoubtedly brought speculation to a fever height as these digital assistants can go out and do things online for you like booking tickets and organizing calendars. These are all things that people have thought that GPT-5 would bring to chatgpt.
GPT-5-Reauting-Alpha-2025-07-13H/T @swishfever pic.twitter.com/HQ5CSD0IC6July 19, 2025
GPT-5 future
Feedback from people using chatgpt agents can actually be used to complete the training for GPT-5. And it needs lots of training if the rumors of the Million-token context window are true. The same goes for the allegedly overall nature of GPT-5.
That would mean that the GPT-5 does not just include switch between features such as visual analysis, coding interpretation and the same abilities as Chatgpt agent; It acts as a unique system. You can ask it to interpret an image, send an e -mail, schedule a meeting and compose and perform a vocal overview from a single prompt.
For example, a parent could coordinate school plans, meals and last minute birthday party logistics at once, or you can plan a trip, book hotels, put it in your calendar and email your family details from a single request. The GPT-5 is reportedly designed to tackle the questions about hallucination and nuanced misunderstanding, allowing people to rely on it more than GPT-4 or GPT-4.5.
There is also the question of memory. Openai has calmly rolled out long-term memory in chatgpt, which means the model remembers things about you, but GPT-5 may make it even more powerful.
Of course, the security questions are already bubbling up. The mention of the GPT-5, tested in biosecurity contexts, have people spooked. If it can resonate about biology well enough to help with complex research, can it also spit out dangerous information if asked for the wrong way? Openai has promised to build protective measures, but the story tells us that people are very good at finding smart ways around digital fences.
Either way, we will all find out soon enough, although it is based on previous launches, it is probably limited to subscribers at the higher levels of chatgpt at first. But you can bet that Openai does not release GPT-5 until they are sure it will not bother them on the first day.
