OpenAI Unveils GPT-5.4, Boosting AI Capabilities for Professional Tasks

SAN FRANCISCO (Azat TV) – OpenAI has announced the release of GPT-5.4, its latest frontier model designed to significantly enhance professional work across various domains. Available in ChatGPT, the API, and Codex, GPT-5.4 integrates advancements in reasoning, coding, and agentic workflows, aiming to deliver more accurate, efficient, and effective results with reduced user interaction.

GPT-5.4 Enhances Professional Productivity

The new model, GPT-5.4, combines the coding prowess of GPT-5.3-Codex with improved functionalities for working across tools, software environments, and professional tasks involving documents, spreadsheets, and presentations. According to OpenAI, this integration leads to complex real-world tasks being completed with greater accuracy and efficiency. In ChatGPT, GPT-5.4 Thinking offers users an upfront plan of its thought process, allowing for mid-response adjustments to better align the final output with user needs. This version also improves deep web research capabilities, especially for highly specific queries, and maintains context over longer thinking processes, resulting in higher-quality answers delivered faster.

Advanced Agentic Capabilities and Token Efficiency

For the Codex and API, GPT-5.4 marks a significant leap as the first general-purpose model with native, state-of-the-art computer-use capabilities. This enables agents to operate computers and execute complex workflows across applications. With support for up to 1 million tokens of context, GPT-5.4 can plan, execute, and verify tasks over extended periods. The model also features improved tool search, helping agents find and utilize the right tools more efficiently. OpenAI highlighted GPT-5.4’s enhanced token efficiency, stating it uses significantly fewer tokens to solve problems compared to GPT-5.2, translating to reduced costs and faster speeds. Benchmark results from OpenAI indicate GPT-5.4’s superior performance on tasks like GDPval (83.0% win rate), SWE-Bench Pro (57.7%), OSWorld-Verified (75.0%), Toolathlon (54.6%), and BrowseComp (82.7%), surpassing its predecessors.

GPT-5.4 Pro and Thinking Variants

GPT-5.4 is being released in two specialized variants: GPT-5.4 Thinking, focused on reasoning, and GPT-5.4 Pro, optimized for maximum performance on complex tasks. GPT-5.4 Thinking is available to all paid ChatGPT subscribers, while GPT-5.4 Pro is reserved for ChatGPT Pro and Enterprise plan users. The API version offers a context window of up to 1 million tokens, the largest ever provided by OpenAI. This advancement is expected to enable more reliable agents, faster developer workflows, and higher-quality outputs across OpenAI’s platforms. Tech analysts note that the model’s performance on benchmarks like Mercor’s APEX-Agents, designed for law and finance skills, shows its potential in specialized professional fields.

New Benchmark Performance and Future Implications

OpenAI shared benchmark results demonstrating GPT-5.4’s improvements. On the GDPval test for knowledge work, GPT-5.4 achieved a record 83% score, surpassing GPT-5.2’s 70.9%. For tasks involving persistent web browsing to find information, GPT-5.4 showed a 17% absolute improvement over GPT-5.2 on the BrowseComp benchmark, with GPT-5.4 Pro reaching 89.3%. In desktop navigation tests using screenshots and mouse/keyboard actions (OSWorld-Verified), GPT-5.4 achieved a 75.0% success rate compared to GPT-5.2’s 47.3%. OpenAI’s release signals a continued push towards more capable AI assistants that can handle intricate professional workflows, potentially reshaping how knowledge work is performed.

The introduction of GPT-5.4 with native computer use capabilities and a 1 million token context window represents a significant step towards more autonomous and integrated AI agents, capable of performing complex, multi-step tasks across various software applications.