Quick Read
- GPT-5.4 is the new flagship model, enhancing AI for professional tasks across API, Codex, and ChatGPT.
- It completes complex tasks with less back-and-forth and significantly fewer tokens than GPT-5.2.
- Achieves a new state-of-the-art on GDPval, matching or exceeding industry professionals in 83% of comparisons.
- Introduces native computer use (CUA) and improved vision, scoring 75% on OSWorld-Verified.
- Features a 1-million-token context window and enhanced tool search capabilities for longer coding tasks and better external tool integration.
The highly anticipated GPT-5.4 model has been officially released, marking a significant leap forward in artificial intelligence capabilities for professional work across its API, Codex, and ChatGPT platforms. The developers announced that this new iteration is their most capable and efficient model to date, designed to tackle complex tasks with reduced user interaction and notably fewer tokens compared to its predecessor, GPT-5.2.
This latest version integrates the industry-leading coding prowess of GPT-5.3-Codex with enhanced performance on long-running tasks, improved front-end user interface generation, and built-in computer use (CUA). GPT-5.4 is engineered to operate seamlessly across various tools, enabling it to analyze and create spreadsheets, presentations, and other professional documents with greater autonomy and precision.
Enhanced Capabilities for Knowledge Work and Efficiency with GPT-5.4
GPT-5.4 delivers more consistent and refined outcomes for real-world professional tasks. In benchmark tests, specifically on GDPval, GPT-5.4 has achieved a new state-of-the-art performance, matching or surpassing industry professionals in 83% of comparisons. This represents a substantial improvement over GPT-5.2, which achieved 71% in the same metrics, underscoring its enhanced reliability and accuracy in knowledge-intensive applications.
A core focus of GPT-5.4’s development was improving efficiency. The model is designed to complete complex tasks with less back-and-forth, requiring significantly fewer tokens than GPT-5.2. This efficiency gain translates into faster processing and potentially lower operational costs for businesses and developers utilizing the API.
Revolutionizing Coding and Computer Use with GPT-5.4
GPT-5.4 introduces native computer use, achieving a state-of-the-art score of 75% on OSWorld-Verified, a dramatic increase from GPT-5.2’s 37.9%. This makes GPT-5.4 the most robust model yet for constructing computer-operating agents, allowing AI to interact with operating systems and applications more effectively. Furthermore, its vision capabilities have been greatly improved, particularly in understanding and processing visual information with higher fidelity.
For developers, GPT-5.4 is positioned as the best coding model released to date. It matches or outperforms GPT-5.3-Codex on SWE-Bench Pro while offering lower latency. A significant enhancement for coding tasks is its larger, 1-million-token context window. This expanded capacity enables the model to handle longer-running and more intricate coding projects, allowing it to use tools, iterate on code, and verify its work within a single, extended session.
Seamless Tool Integration and Advanced Web Search in GPT-5.4
The new model significantly advances how AI interacts with external tools. GPT-5.4 introduces tool search functionality, which allows models to dynamically load necessary tools into their context without disrupting the cache. This innovation improves tool calling performance across various reasoning efforts, particularly benefiting low-latency applications where quick, adaptive tool use is critical.
GPT-5.4 also boasts improved agentic web search capabilities. On BrowseComp, a benchmark designed to measure an AI agent’s ability to persistently browse the web for hard-to-locate information, GPT-5.4 achieved a 17 percentage point leap over GPT-5.2. The professional version, GPT-5.4 Pro, sets a new state-of-the-art record of 89.3%, indicating a profound improvement in the model’s ability to conduct sophisticated and persistent online research.
The release of GPT-5.4 signifies a strategic move towards more autonomous and integrated AI systems, emphasizing not just raw processing power but also the ability to interact with diverse digital environments and tools. Its advancements in computer use and tool integration are likely to accelerate the development of highly capable AI agents, transforming professional workflows across numerous industries by enabling more complex, multi-step tasks to be handled with greater efficiency and less human oversight.