GPT-5.4: OpenAI Redefines Professional AI Standards

Launch of GPT-5.4: A Major Advancement for ChatGPT

OpenAI recently announced the launch of GPT-5.4, an enhanced version of its artificial intelligence model, now available in ChatGPT under the name GPT-5.4 Thinking, as well as in the API and Codex. This model represents the pinnacle of current technology for professional applications. In parallel, OpenAI introduced GPT-5.4 Pro, aimed at users looking to maximize performance on particularly complex tasks.

The GPT-5.4 model incorporates the latest advancements in reasoning, programming, and agentic workflows. It builds on the advanced programming capabilities of GPT-5.3-Codex, while improving interaction with various tools and software environments. This includes professional tasks involving spreadsheets, presentations, and documents. The result is a model capable of performing complex work with precision and efficiency, reducing the need for multiple exchanges to achieve the desired outcome.

Improvements in ChatGPT

In ChatGPT, GPT-5.4 Thinking now offers a feature that allows users to visualize its reasoning in real-time. This gives users the ability to adjust the direction of the response as it is being developed, ensuring a final result that aligns more closely with their expectations, without requiring new interactions. Additionally, GPT-5.4 Thinking enhances deep web searching, particularly for very specific queries, while maintaining context for questions that require extended reasoning. These improvements translate into higher quality, faster, and more relevant responses.

Extended Capabilities in Codex and the API

In Codex and the API, GPT-5.4 stands out as the first general-purpose model with native and advanced computer usage capabilities. This allows agents to execute complex workflows across different applications. The model supports up to 1 million tokens of context, enabling planning, execution, and verification of tasks over long periods. GPT-5.4 also improves the efficiency of models within vast ecosystems of tools and connectors, thanks to optimized tool searching, helping agents find and use the right tools without compromising intelligence.

GPT-5.4 uses significantly fewer tokens to solve problems compared to GPT-5.2, resulting in reduced token usage and higher speeds.

Impressive Comparative Performance

The performance of GPT-5.4 stands out in several benchmarks. For example, in the GDPval benchmark, which evaluates agents' ability to produce well-defined intellectual work across 44 professions, GPT-5.4 surpasses or matches professionals in 83% of cases, compared to 71% for GPT-5.2. This benchmark covers concrete tasks from 9 major sectors of the U.S. economy, requiring deliverables such as business presentations or emergency care plans.

In other benchmarks, GPT-5.4 shows impressive performance: SWE-Bench Pro at 57.7%, OSWorld-Verified at 75.0%, Toolathlon at 54.6%, and BrowseComp at 82.7%.

Focus on Intellectual Work

Building on the reasoning capabilities of GPT-5.2, GPT-5.4 delivers even more consistent and polished results for critical professional tasks. In an internal benchmark on spreadsheet modeling tasks, GPT-5.4 achieved an average score of 87.5%, compared to 68.4% for GPT-5.2. Furthermore, for presentations, human evaluators preferred those from GPT-5.4 68% of the time, thanks to a more refined aesthetic and more effective use of image generation.