OpenAI GPT-5.4 发布 — 1M 上下文、计算机使用功能，以及 GPT-4o 到 GPT-5 的大规模退役

OpenAI officially launched GPT-5.4 on March 5, offering three variants: the basic model GPT-5.4, the GPT-5.4 Thinking which explicitly shows reasoning processes, and the high-performance GPT-5.4 Pro. Key specifications include a 1M token context window capable of processing approximately 3,000 pages at once, native computer-use functionality allowing the model to directly manipulate screens and mouse/keyboard. Additionally, GPT-5.4 achieved an OSWorld-Verified score of 75.0%, surpassing the human benchmark of 72.4%. This version integrates reasoning, coding, and agentic workflows into a single model, replacing the previous separation of GPT-5.3-Codex, o-series, and GPT-5.x.