OpenAI launches GPT-5.4, a high-performance and efficient model, available in three variants

OpenAI has announced the launch of GPT-5.4, a base model intended for professional use, described as the most capable and efficient frontier model of the company. It comes in three variants: the standard GPT-5.4 model, GPT-5.4 Thinking for complex reasoning tasks, and GPT-5.4 Pro, optimized for high-performance tasks. A notable feature is the extended context window, which can reach 1 million tokens, allowing for the processing of large amounts of information in single requests. OpenAI also reports improved token efficiency, reducing operational costs for AI-based applications. Benchmark performance has increased, with GPT-5.4 achieving record results in computerized interaction tests and professional capacity assessments in fields such as law and finance. The model is also more reliable, having 33% fewer factual errors compared to the previous version. OpenAI has introduced a new system called Tool Search, which enhances interaction with external tools, reducing token usage. Safety assessments have been expanded, and initial tests suggest that the Thinking version of the model has a limited capacity to intentionally manipulate reasoning explanations. With these improvements, GPT-5.4 represents an important step towards AI systems capable of managing complex professional workflows at scale.