OpenAI is Testing an Improved Model for Code Generation — o3-Alpha is Already on WebArena

OpenAI has begun testing a new version of its model, o3-alpha, on the WebArena platform, where users can directly compare the performance of different AI systems. The model is presented under the name “Anonymous-Chatbot-0717” and, according to BleepingComputer, significantly outperformed previous versions in tasks related to coding and web page creation.

Model Features

O3-alpha can handle even unstructured and vague queries, easily creating HTML pages, basic UI layouts, and even simple web games. This makes the model particularly useful for novice programmers, front-end developers, and interface designers.

According to an analysis of the WebArena platform code, the current test version of the model is recorded as modelApiId: “o3-alpha-responses-2025-07-17”.

More Details

Despite active testing, o3-alpha is not yet a publicly available model and is not the open-source release that OpenAI previously mentioned. It is also important to note that the model called GPT-4.5 or GPT-4.1 does not appear in this case; this is an intermediate version, probably intended for internal optimization before the launch of GPT-5.

Although OpenAI has not officially commented on the details, the improvements observed in o3-alpha, especially in code generation, logic, and UX solutions, may indicate that this model already includes technologies that will be part of the future GPT-5.

GPT-5 is expected to combine advanced logic, multimodal capabilities (working with images, sound, and video), and improved interaction algorithms.

Against this backdrop, o3-alpha can be seen as a transitional stage to the next generation of universal AI.