OpenAI launches GPT-5 with automatic mode selection and improved logic

      OpenAI has launched GPT-5, the fastest and most logical neural network in its history. In the new version, users no longer select the operating mode manually. The model automatically determines when a quick response is needed and when a deep analysis is required. On the same day OpenAI introduced its development, Elon Musk with Grok 4 Heavy announced a new round of competition in the artificial intelligence market. The question of leadership has once again become open.

      GPT-5 is an updated artificial intelligence model that autonomously chooses how to process requests. The speed or detail of the response depends on the user's task. Manual switching between versions is now irrelevant: all calculations are performed by the neural network itself. Extended settings and increased limits are available for Plus and Pro subscribers. The free plan also allows trying GPT-5, albeit with some limitations.

      Significant changes concern speed and quality of operation. GPT-5 responds faster and constructs logical chains more reliably. The number of errors and random "hallucinations" has nearly halved compared to GPT-4. In Thinking mode, the model makes mistakes even less often—about 80% less than before. This result allows GPT-5 to be used for real tasks where stability and predictability are crucial.

      The context window has been increased to 256,000 tokens. The model easily handles large documents, long series of clarifications, and tasks involving multimedia content. GPT-5 understands text, images, videos, and audio. It now has agent functions: the model can utilize a browser and API to retrieve additional information or perform external actions. In programming, GPT-5 demonstrates high performance. On the SWE-bench Verified test, the model scored 74.9%, and on Aider Polyglot—88%. It is capable of creating web applications, working with Python projects in parts, analyzing errors from screenshots, and solving complex development tasks.

      For business and science, this is a significant update. GPT-5 supports current APIs and tools, can create its own LLM agents, and integrates with Microsoft products. Pro users now have access to additional communication styles, called "personas."

      Competitors have not been idle either. Elon Musk reminded about his Grok 4 Heavy model, which, according to him, already surpassed GPT-5 in the ARC-AGI test. This test assesses AI's ability to think like humans. Grok 4 Heavy scored around 68%, GPT-5—65.7%. Grok has its features: high cost, lower speed, and niche applications. Unlike it, GPT-5 is designed for the mass user.

      OpenAI's presentation also raised questions. In comparative charts, some experts noted overly optimistic estimates and overly bright data. However, according to independent test results, GPT-5 currently holds leading positions, especially in tasks requiring comprehensive analysis and integration with other services.

      In practical applications, GPT-5 will become a reliable tool for those who value predictable and understandable AI. The model responds quickly and accurately, without unnecessary "creativity." Grok 4 Heavy currently outperforms in complex intelligence tests but remains a product for specific tasks and enthusiasts. The competition in the AI market continues. The leader will be whoever makes technologies more accessible and easier to use.