OpenAI launches GPT-5.4 with improved reasoning, coding, and computer use capabilities

OpenAI has launched GPT-5.4, a new frontier model designed for professional workloads, combining advanced reasoning, coding, and agent-based workflows into a single system.

The model is rolling out across ChatGPT, the API, and Codex, with a higher-performance GPT-5.4 Pro version also available for complex tasks requiring maximum compute and accuracy.

GPT-5.4 integrates the coding capabilities introduced in GPT-5.3 Codex while improving performance across tools, software environments, and professional tasks such as spreadsheets, presentations, and document creation. The model can also outline its reasoning plan in advance within ChatGPT, allowing users to guide responses mid-process.

The release introduces native computer use capabilities, enabling AI agents to interact with operating systems, websites, and applications through mouse, keyboard, and visual inputs. Developers can use the model to automate multi-step workflows across software environments.

GPT-5.4 supports context windows of up to one million tokens and introduces tool search, allowing AI systems to dynamically locate and use external tools without loading all tool definitions into memory. This reduces token usage while improving performance in complex workflows.

Benchmarks show improvements across multiple tasks. On the GDPval benchmark for professional knowledge work, GPT-5.4 matches or exceeds industry professionals in 83% of comparisons, compared with about 71% for GPT-5.2. The model also outperforms earlier versions on coding tests, web browsing tasks, and computer use benchmarks.

The model is also designed to be more efficient. OpenAI says GPT-5.4 requires fewer tokens to solve many reasoning tasks compared with GPT-5.2, improving response speed and reducing costs for developers.

GPT-5.4 Thinking is now available to ChatGPT Plus, Team, and Pro users, while GPT-5.4 Pro is available for Pro and Enterprise plans. Developers can access both models through the API.