We chose OpenAI for reliability, API maturity, and how well it fit our multi-step research workflow. It helped us handle extraction, structuring, and fast reasoning steps consistently, while still letting us combine it with other models where they were stronger.