Another big earthquake in the world of technology! OpenAI, considered an artificial intelligence giant, has officially presented its newest and most powerful model - GPT-5.4. This new system has become not just an "interlocutor," but a real digital expert, solving complex technical problems.
Zamin.uz analyzed the most interesting aspects of this technological revolution.
The new generation model is specially sharpened to perform professional and technological tasks. It combines strong logical reasoning, the ability to write program code, and the ability to automate daily work processes.
Three different versions for three different needs
OpenAI divided the model into three main areas based on user needs:
- Standard version: Mainly for developers and system binders via API.
- GPT-5.4 Thinking: for constructing and deeply analyzing complex logic chains on the ChatGPT platform.
- GPT-5.4 Pro: A "super-model" designed for heavy tasks requiring the highest computing power.
Record results and practical efficiency
The new model surpassed its competitors not only in theory but also in practical tests. She recorded record scores on the OSWorld-Verified and WebArena Verified control exams. This means that AI can now work freely with various software and online tools just like humans.
Economic and qualitative advantages: According to the head of Mercor, Brendan Foudy, GPT-5.4 provides not only high-quality services in the creation of financial models, preparation of presentations, and analysis of legal documents, but also much faster and cheaper than other models.
Computer Management: The Future of AI Agents
The most striking feature of the new model is its ability to directly control the computer. With the help of the Codex and updated API, AI agents can perform complex operations between various applications. Most importantly, it doesn't lose context during the process.
GPT-5.2
- Accuracy of facts - standard;
- Logical errors - exist;
- Computer control - limited.
GPT-5.4
- Accuracy in facts - 33% higher;
- Logical errors - significantly less;
- Computer control - extended.
Has "deception" been stopped?
What worries people most is AI's deliveries of false information (hallucinations). In the GPT-5.4 Thinking version, a serious solution to this problem was found. The new model makes 33% fewer errors on individual facts compared to previous versions.
Another convenience: Now ChatGPT will show you its work plan before responding. This allows you to correct the direction of the response from the very beginning and save time.
In your opinion, how much does artificial intelligence's direct control of the computer make our workflow easier, or is it dangerous from a security perspective?
Read “Zamin” on Telegram!
