OpenAI on March 5, 2026, released GPT-5.4, its most capable and efficient frontier model to date, combining advanced reasoning, coding, and agentic workflows into a single unified system.
The model is rolling out across ChatGPT (as GPT-5.4 Thinking), the API, and Codex, with a higher-performance GPT-5.4 Pro variant available for users requiring maximum compute on complex tasks.
GPT-5.4 consolidates capabilities previously spread across separate models, integrating the industry-leading coding strengths of GPT-5.3-Codex with improved general reasoning and native computer-use capabilities.
The result is a model engineered for end-to-end professional workflows from spreadsheets and presentations to complex multi-step agentic tasks with less back-and-forth interaction required from users.
In ChatGPT, GPT-5.4 Thinking introduces an upfront reasoning plan that allows users to interrupt and redirect the model mid-response without restarting, enabling more targeted, context-accurate outputs. This real-time steerability is a notable shift from prior reasoning models, where course corrections required starting over entirely.
GPT-5.4 Launched
GPT-5.4 sets new state-of-the-art scores across several critical industry benchmarks:
BenchmarkGPT-5.4GPT-5.3-CodexGPT-5.2GDPval (wins or ties)83.0%70.9%70.9%SWE-Bench Pro (Public)57.7%56.8%55.6%OSWorld-Verified75.0%74.0%47.3%Toolathlon54.6%51.9%46.3%BrowseComp82.7%77.3%65.8%
On GDPval, which tests agents across 44 occupations spanning the top 9 U.S. GDP industries, GPT-5.4 matches or exceeds industry professionals in 83% of comparisons, up from 70.9% with GPT-5.2.
On the BigLaw Bench evaluation for legal document work, the model scored 91%, according to Harvey’s Head of Applied Research, Niko Grupen.
GPT-5.4 is OpenAI’s first general-purpose model with native computer-use capabilities, enabling agents to interact directly with software through screenshots, mouse commands, and keyboard inputs.
On OSWorld-Verified, it achieves a 75.0% success rate, surpassing human performance benchmarked at 72.4% and far exceeding GPT-5.2’s 47.3%.
On WebArena-Verified, GPT-5.4 achieves a 67.3% browser success rate, while scoring 92.8% on Online-Mind2Web using screenshot-based observations alone.
The model also supports 1 million tokens of context in the API, enabling long-horizon task execution across large-scale agent workflows matching context window offerings from Google and Anthropic.
OpenAI emphasized that GPT-5.4 is its most factual model yet, with individual claims 33% less likely to be false and full responses 18% less likely to contain errors compared to GPT-5.2.
The model also delivers significant token-efficiency gains, using substantially fewer tokens to solve the same reasoning problems, translating directly into reduced API costs and faster response times for enterprise developers.
In production environments, Mainstay CEO Dod Fraser reported GPT-5.4 achieved a 95% first-attempt success rate across ~30,000 property portals, completing sessions three times faster while using 70% fewer tokens versus prior computer-use models.
GPT-5.4 Thinking is available now for ChatGPT Plus, Team, and Pro subscribers, replacing GPT-5.2 Thinking over the next three months. Developers can access GPT-5.4 and GPT-5.4 Pro through the OpenAI API, with priority processing enabled for faster token velocity in production environments.
Follow us on Google News, LinkedIn, and X for daily cybersecurity updates. Contact us to feature your stories.
The post OpenAI Launches GPT-5.4 With Advanced Reasoning, Coding, and Computer-Use Capabilities appeared first on Cyber Security News.
