OpenAI announced the rollout of new GPT-5.4 Thinking and GPT-5.4 Pro versions in ChatGPT. Additionally, the GPT-5.4 model has become accessible to developers via API and the Codex platform. This updated version focuses on several key areas of development: enhanced logical reasoning, expanded programming capabilities, and support for scenarios where agent actions are crucial.
Key points to note:
– GPT-5.4 outperforms all others in tasks related to agent activity
– Significant improvements in mathematics performance
– Achieves impressive results in scientific research (e.g., GPQA)
– Substantially better handling of web interfaces and tools
Essentially, this model was designed specifically for agent-based interactions and complex reasoning — it’s not limited to simple chatbots.
According to company representatives, GPT-5.4 is considered one of the most accurate and efficient models to date. It uses fewer tokens to complete tasks and operates faster compared to previous versions.
In the updated ChatGPT version, the GPT-5.4 Thinking model received an improved reflection mode: with longer analysis, it better retains context and can conduct in-depth internet research.
A new feature for managing the response process has also been introduced. Users can now stop text generation at any moment and modify the task by adding additional instructions.
This feature has already begun appearing in the web version of ChatGPT and on Android, with iOS support expected shortly thereafter.
Created with n8n:
https://cutt.ly/n8n
Created with syllaby:
https://cutt.ly/syllaby