OpenAI Unveils GPT-5.4: A Frontier Model with Native Computer Use, 1M Token Context, and Breakthrough Coding Capabilities
OpenAI has officially launched its latest frontier model, GPT-5.4, which has been simultaneously deployed across ChatGPT, API, and Codex platforms. As currently the most powerful specialized work model, GPT-5.4 deeply integrates cutting-edge reasoning, coding, and agent workflow technologies, while introducing native computer operation capabilities into general-purpose models for the first time. In terms of core functionality, GPT-5.4 significantly enhances efficiency in both coding and computer operations. Building upon the top-tier coding capabilities inherited from GPT-5.3-Codex, it can more precisely control software systems via mouse/keyboard commands and repository interactions, achieving record scores on OSWorld-Verified and WebArena-Verified benchmarks. Regarding document processing, the new model demonstrates substantially improved performance in creating Excel spreadsheets and PowerPoint presentations as well as parsing documents, further reducing hallucination rates and delivering greater factual accuracy. Additionally, tool invocation efficiency has been optimized by effectively addressing context expansion issues during tool calls, markedly lowering token consumption and improving response speeds. For professional users, OpenAI concurrently released the GPT-5.4 Pro edition, specifically designed for handling extremely complex tasks. Within ChatGPT, users can experience the GPT-5.4 Thinking mode, which supports extended reasoning processes alongside enhanced long-text comprehension and deep web research capabilities, enabling more precise identification within vast information sets. Regarding performance metrics, GPT-5.4 excelled across multiple benchmark tests: scoring 83.0% on GDPval (general knowledge work), reaching 57.7% on SWE-Bench Pro (advanced programming), and leading with an impressive 82.7% score on the browsing task BrowseComp. On the API front, GPT-5.4 supports a context window of up to one million tokens; although the unit price per token has increased slightly, overall task costs and latency have decreased significantly due to higher execution efficiency. Currently, GPT-5.4 is gradually rolling out to ChatGPT Plus, Team, Pro, and enterprise users, with full API access now available.
