🚨OpenAI released GPT-5.4 on March 5, 2026. It is their most capable frontier model so far, designed specifically for professional work. This version combines advanced reasoning, high-level coding capabilities, agentic workflows, and native computer use for the first time in a mainline OpenAI model.The biggest new feature is native computer use through the API and Codex. The model can view screenshots, interpret what it sees, move the mouse, type commands, and navigate any desktop application to complete multi-step tasks automatically. It scored 75.0% on the OSWorld-Verified benchmark, slightly above the human average of 72.4%.It has a 1 million token context window in the API and Codex, allowing it to process entire large codebases, full documents, massive spreadsheets, or long agent histories in a single prompt without losing context. Tool integration is smarter with a new tool search feature that reduces token usage by about 47% on complex workflows such as handling refunds, support tickets, or multi-app automation. Multimodal performance is improved for better document parsing, image and video understanding, and visual reasoning on charts and presentations.Reliability is significantly better. It is OpenAI's most factual model to date, with 18% fewer errors and 33% fewer hallucinations compared to GPT-5.2. In Thinking mode it shows a reasoning plan upfront so users can interrupt and steer the response mid-generation. Inference is faster in optimized modes with lower overall token consumption for high-quality outputs.On professional benchmarks it matches or exceeds human expert performance in roughly 83% of knowledge-work tasks across 44 occupations. It scores around 83% on GDPval for expert knowledge work, up to 89.3% on BrowseComp for persistent web navigation, and about 87% on junior-level spreadsheet modeling. It often outperforms recent Claude models in creative writing evaluations and produces cleaner, more complete code with stronger handling of edge cases in agent loops.Versions include GPT-5.4 Standard as the balanced default, GPT-5.4 Pro for maximum performance on the hardest tasks (available to Pro and Enterprise users), and GPT-5.4 Thinking mode in ChatGPT which replaces the previous Thinking version for paid tiers. It is rolling out gradually to Plus, Team, Pro, Enterprise, and Edu users in ChatGPT, while the API is already live with 1M context support. Codex has full integration for coding agents and large-context automation.Compared to GPT-5.2 it is much more efficient, factual, and practical for real work. It sets a new standard for native computer control and pushes closer to fully autonomous professional AI agents.