Reports have emerged in the AI community about a new version of GPT-5, noticed in the Biosec Benchmark repository. The mention of “GPT-5 Reasoning Alpha” is dated July 13, indicating internal use of the model for research in the field of biosecurity. Biosec Benchmark evaluates the ability of models to identify and prevent risks from laboratory viruses, so testing GPT-5 in this context highlights its role in important scientific tasks.
Soon after this appearance, another new model was recorded online — o3 Alpha Responses, which on July 17 showed high results on the Web Arena platform. It stands out for its rapid creation of complex web applications, such as games like Minecraft, and demonstrates advanced skills in generating SVG graphics. It is not yet known whether this model is related to GPT-5, but its performance noticeably exceeds previous public samples.
These events coincided with comments from OpenAI employees, who confirmed that the release of GPT-5 is approaching. At the same time, they emphasized that models with the level of mathematical olympiads are still in the experimental stage and will not appear in the coming months. Thus, OpenAI continues testing and improving the new model, paying attention to both research tasks and practical opportunities for programmers.
Currently, research and development communities are closely monitoring the situation, expecting further updates. According to the latest signals, the main focus is on expanding the model’s capabilities in logical reasoning and applied tasks.
