OpenAI has introduced the new GPT-5.4 model. This model supports context up to one million tokens and can outperform humans in navigating operating system interfaces. Reports rozetked.me.
GPT-5.4 also includes GPT-5.4 Pro and GPT-5.4 Thinking versions. This model combines logical reasoning, programming, and independent agent processes. OpenAI's new model features a natural computer control function, achieving 75% success in this area.
This model has shown significantly higher results compared to the previous GPT-5.2 version (47.3%) and the average human result (72.4%). In the SWE-Bench Pro programming benchmark, GPT-5.4 outperforms the GPT-5.3-Codex model and operates with less latency.
The new model demonstrates expert-level or higher results in 83% of cases when performing real work tasks. Additionally, GPT-5.4 supports processing original images up to 10.24 megapixels.
OpenAI has started the rollout of new models in the ChatGPT app and web version, and they are also available via API and Codex.
Read “Zamin” on Telegram!
