OpenAI’s new GPT-5.4 model is rolling out across ChatGPT, the API, and Codex as the company’s latest step toward more capable professional AI systems. The model introduces GPT-5.4 Thinking for ChatGPT users and a more advanced GPT-5.4 Pro option designed for complex workflows.

The release focuses heavily on professional tasks. OpenAI says GPT-5.4 improves reasoning, coding, research, document creation, spreadsheet modeling, and multi-step tool workflows while reducing the number of interactions needed to complete tasks.

A shift toward AI built for professional work

GPT-5.4 is positioned as OpenAI’s most capable model for knowledge work and technical productivity. The company says the system combines the reasoning improvements from earlier models with the coding capabilities previously introduced in GPT-5.3-Codex.

The model is designed to handle tasks commonly performed by professionals. These include building spreadsheets, creating slide presentations, writing reports, analyzing documents, and managing multi-step workflows across different software tools.

According to OpenAI’s benchmark testing, GPT-5.4 achieved stronger results than previous versions across several industry evaluations. On GDPval, which measures performance across professional occupations, the model matched or exceeded expert-level outputs in 83 percent of comparisons.

The company also reported improvements in coding performance. GPT-5.4 scored 57.7 percent on SWE-Bench Pro, a benchmark used to measure software engineering capabilities, slightly ahead of previous model versions.

OpenAI says the model is also more efficient when reasoning through complex problems. It reportedly uses fewer tokens than GPT-5.2 to complete many tasks, allowing faster responses while maintaining accuracy.

Computer control and longer context capabilities

A major technical change in GPT-5.4 is its ability to interact directly with computer environments. The model can analyze screenshots, control user interfaces, and automate actions through keyboard and mouse instructions.

This feature allows AI agents built with the model to complete real workflows such as navigating websites, processing documents, scheduling events, or managing business tools.

GPT-5.4 also introduces much larger context capabilities. In some environments, the model can handle up to one million tokens of information, enabling long-running tasks that require large datasets or extended documents.

The system also improves how AI agents choose and use tools. Instead of loading every available tool into the prompt at once, GPT-5.4 can search for the relevant tool when needed, reducing token usage and improving efficiency in large tool ecosystems.

Changes coming to ChatGPT users

In ChatGPT, GPT-5.4 Thinking is replacing GPT-5.2 Thinking for paid plans including Plus, Team, and Pro. OpenAI says the earlier model will remain available in a legacy section for a limited period before being retired.

The company also introduced GPT-5.4 Pro for users who need maximum performance on complex tasks such as software development, financial modeling, or technical research.

Developers can access GPT-5.4 through the OpenAI API, with pricing slightly higher than the previous generation. The company says the increased capability and token efficiency are intended to offset the higher cost per token for many workloads.

Alongside the new model, OpenAI is expanding tools that allow AI systems to work across spreadsheets, documents, and software environments. These updates are part of the company’s broader effort to move AI from conversational assistance toward task completion and automation.

OpenAI’s new GPT-5.4 model represents a step toward more capable AI systems built for real work rather than simple conversation. As the rollout expands across ChatGPT and developer tools, the model will likely play a central role in how businesses and professionals integrate AI into everyday workflows.

FYI (keeping you in the loop)-