GPT-5.4

OpenAI Launches GPT-5.4, First AI to Outperform Humans at Computer Control

What Happened On Thursday, March 5, 2026, OpenAI announced the release of GPT-5.4, available in three versions: the standard model, GPT-5.4 Thinking (with enhanced reasoning capabilities), and GPT-5.4 Pro (high-performance version). The release represents what OpenAI calls “our most capable and efficient frontier model for professional work.” The standout achievement is GPT-5.4’s performance on the OSWorld-Verified benchmark, where it scored 75% compared to human performance of 72.4%. This benchmark tests a model’s ability to navigate desktop environments using only screenshots and keyboard/mouse actions, essentially measuring how well AI can operate a computer like a human would.

March 7, 2026 Read more →

OpenAI Launches GPT-5.4 with Computer Control Capabilities

What Happened OpenAI officially released GPT-5.4, positioning it as their most capable model yet for autonomous computer operation. Unlike previous AI models that could only respond to text or generate content, GPT-5.4 can actively control a computer interface, clicking buttons, navigating applications, and completing multi-step tasks across different software programs. The new model builds on existing GPT capabilities while adding what OpenAI calls “native computer use” - the ability to see, understand, and interact with computer interfaces just like a human user would.

March 5, 2026 Read more →