GPT-6 Astra now handles long-context legal drafting well enough to produce structured, matter-aware documents in one pass rather than requiring heavy retrieval scaffolding and chunk-stitching. If you're building document-generation pipelines, this shifts effort away from managing context windows and RAG plumbing toward validation and citation-checking of longer coherent outputs—but expect higher per-call token costs and latency for those larger contexts.