OpenAI Unveils GPT-5.4 With Million-Token Context and Desktop Autonomy
What Happened
OpenAI introduced GPT-5.4 with a one-million-token context window and the ability to operate desktop software end-to-end, posting a 75% score on the OSWorld-V benchmark for multi-step productivity tasks. The model is positioned for enterprise agent deployment and unlocks workflows requiring entire codebases, contract sets, or research libraries in a single prompt.
My Take
OSWorld-V at 75% is the number that matters here, not the context window. A model that can drive your accounting software, your CRM, and your filesystem reliably is no longer an assistant — it's a junior employee with no salary, no PTO, and no learning curve between hires. The "human in the loop" framing will quietly disappear from enterprise pitches by fall. Companies should start writing SOPs as model-readable specs, because the operator using them in 2027 won't be human.
Read Original Source