Introducing GPT-5.4

OpenAI announced the release of GPT-5.4 and GPT-5.4 Pro across ChatGPT, the API, and Codex. GPT-5.4 integrates recent advances in reasoning, coding, and agentic workflows, incorporating the coding prowess of GPT-5.3-Codex while improving performance across tools, software environments, and professional documents like spreadsheets and presentations. In ChatGPT, GPT-5.4 Thinking offers upfront planning for user adjustments mid-response and improves deep web research. For the API and Codex, GPT-5.4 is the first general-purpose model with native, state-of-the-art computer-use capabilities, supporting up to 1M tokens of context and featuring tool search for efficient tool use. Benchmarks show GPT-5.4 achieving new state-of-the-art results in knowledge work (83.0% on GDPval), computer use (75.0% on OSWorld-Verified, surpassing human performance), and coding (matching or outperforming GPT-5.3-Codex on SWE-Bench Pro). The model is also more factual, with 33% fewer false individual claims than GPT-5.2. Safety measures treat GPT-5.4 as High cyber capability, and new features like tool search significantly reduce token usage and cost in tool-heavy workflows. GPT-5.4 Thinking in ChatGPT now provides a preamble for complex queries, allowing users to adjust direction mid-response.

(Source：OpenAI)