OpenAI says GPT-5.5's improvements are strongest in agentic coding, computer use, and early scientific research, which require reasoning across longer contexts (Madison Mills/Axios)