1M-token context windows: is it time to rethink your RAG architecture?
GPT-5.6 and Kimi K3 make 1M-token context the norm. Cost, recall, and hybrid architecture: what should actually change in your RAG.
Long Context