Another breakthrough has been recorded in the world of artificial intelligence. The GPT 5.4 model achieved absolute leadership in Vibe Code Bench v1.1 tests with a result of 67.42 percent. This figure is 5.7 percent higher than the previous record holder, the GPT 5.3 Codex model. The third place was taken by the Claude Opus 4.6 model with a result of 57.57 percent. Reports habr.com.
Unlike simple coding or bug fixing, this test evaluates the model's ability to create an entire web application from scratch. The program must independently carry out the process from an empty folder to a finished service. During the test, complex tasks such as authentication, databases, and payment systems are assigned to the artificial intelligence. Up to five hours are allocated for each project, and the final result is checked by a special agent.
Analysis shows that high accuracy does not always mean economic efficiency. For example, the Claude Opus 4.6 model achieves almost equal results while consuming fewer resources than GPT 5.4. This is an important factor for developers in optimizing costs. Nevertheless, the pace of artificial intelligence development is astonishing: in just half a year, the best performance indicators have doubled.
Currently, GPT 5.4 still makes errors in some tasks. Approximately one-third of the applications created by the model do not work fully. Nevertheless, the technology is expected to fundamentally change the programming process in the future and significantly ease human labor.
Read “Zamin” on Telegram!
