GPT-5.4 Thinking Breakthrough: Surpassing Humans in OSWorld-V with Advanced Inference and 1M Context

This is a monumental leap for autonomous AI agents, showcasing OpenAI's relentless push towards highly capable reasoning models. Achieving a 75% score on the OSWorld-V benchmark to officially surpass the human baseline of 72.4% is a thrilling milestone that signals AI is ready to handle real-world, complex desktop tasks. Furthermore, the introduction of granular Inference controls and a massive 1M Context Window opens up incredible possibilities for developers building the next generation of long-running, self-sufficient digital workers.