GPT-5.4 is a large language model (LLM) developed and released by OpenAI on March 5, 2026. It represents a significant advancement in OpenAI’s Generative Pre-trained Transformer series, aiming to unify and enhance capabilities previously distributed across specialized models. The model is available in two primary variants: GPT-5.4 Thinking, designed for advanced reasoning tasks and available to paid ChatGPT subscribers, and GPT-5.4 Pro, which offers maximum performance for high-end research and enterprise use cases via API. This release consolidates the strengths of its predecessors, specifically merging the coding prowess of GPT-5.3 Codex with the broader reasoning and knowledge work capabilities of GPT-5.2.
A core improvement in GPT-5.4 is its enhanced accuracy, with OpenAI reporting a 33% reduction in factual errors compared to GPT-5.2. It also features native computer use, allowing the model to operate computers through Playwright code and direct mouse/keyboard commands from screenshots, a capability not built into previous general-purpose OpenAI models. This functionality enables GPT-5.4 to perform complex multi-step workflows with greater autonomy. Furthermore, it boasts a substantial context window of up to 1 million tokens, supporting both text and image inputs for high-context reasoning and multimodal analysis. The model also introduces an efficient Tool Search system, which optimizes token usage by retrieving tool definitions only when needed, potentially reducing token usage by up to 47%.
GPT-5.4 is positioned as OpenAI’s most capable and efficient frontier model for professional work, excelling in areas like coding, document understanding, and agentic workflows. It enables features such as upfront thinking plans and mid-task steering, allowing users to review and adjust the model’s approach during a task. For developers, GPT-5.4 effectively replaces the need for a specialist coding model like GPT-5.3 Codex by integrating advanced coding capabilities directly, supporting the entire software development lifecycle from architecture to testing. In the context of large-scale data processing and AI applications, the enhanced capabilities of models like GPT-5.4 make them increasingly relevant for generating embeddings that can be stored and efficiently queried in vector databases such as Milvus. This allows for advanced semantic search, recommendation systems, and anomaly detection over vast datasets, leveraging the model’s improved understanding and context handling.