OpenAI’s GPT-5.4 Thinking: A New Era of Interactive AI

OpenAI has begun rolling out GPT-5.4 in ChatGPT, introducing a new “GPT-5.4 Thinking” mode that allows users to interrupt the model whereas it is generating a response and steer the output before the final answer is produced. This advancement marks a significant step towards more collaborative and controlled AI interactions, particularly for professional workflows and complex tasks.

Interruptible Reasoning: Guiding AI in Real-Time

GPT-5.4 Thinking introduces a planning step where the system outlines its approach to a task. Users can review this plan and adjust the direction before the model generates a final output. This feature addresses a common frustration with AI models – the require for multiple rounds of prompting to achieve the desired result. As Nick Turley, Head of ChatGPT at OpenAI, outlined on LinkedIn, the new mode allows users to “steer the response while it’s still working instead of needing multiple back-and-forth turns.”

Enhanced Capabilities Across Multiple Domains

GPT-5.4 combines recent advances in reasoning, coding and agent-based workflows into a single model. According to OpenAI, the system incorporates coding capabilities developed through earlier Codex models while improving its interaction with documents, spreadsheets, presentations, and other professional tools. The model also boasts improvements in deep web research and long-context reasoning, delivering faster and more focused answers to highly specific questions.

Key Improvements in GPT-5.4

• Coding: Incorporates the industry-leading coding capabilities of GPT-5.3-Codex. OpenAI
• Document Understanding: Improved ability to operate with complex documents. OpenAI
• Tool Leverage: Enhanced interaction with various software environments. OpenAI
• Instruction Following: More accurate and efficient response to user prompts. OpenAI
• Token Efficiency: Uses significantly fewer tokens to solve problems and answer queries, resulting in faster speeds. VICE
• Context Window: Supports a context window of up to 1 million tokens, enabling the analysis of large codebases and document collections. 9to5Mac

Expanded Computer Use and Agent Systems

GPT-5.4 is OpenAI’s first “mainline model” with built-in computer-use capabilities, enabling agents to interact directly with software to complete, verify, and fix tasks in a build-run-verify-fix loop. It is also the first mainline model trained to support compaction, allowing for longer agent trajectories while preserving key context. These features are particularly valuable for agent systems and complex, multi-step workflows.

Availability

GPT-5.4 Thinking is currently available for ChatGPT Plus, Team, and Pro subscribers and will replace GPT-5.2 Thinking within three months. TechRadar The model is also rolling out in OpenAI’s API and Codex.

The Future of AI Interaction

The introduction of GPT-5.4 Thinking reflects a broader shift in AI development, moving beyond simple conversational use cases towards supporting structured knowledge work, software development, and complex professional tasks. By offering users greater control and transparency over the AI’s reasoning process, OpenAI is paving the way for a more collaborative and productive future with artificial intelligence.