Million-Token Context Windows: Rewriting Diligence and Audit

Frontier models now hold an entire deal book, codebase, or audit corpus in a single context window. The teams updating their reference architecture are pulling weeks out of M&A diligence and audit prep.

Frontier models now routinely accept context windows in the 1 to 2 million token range. That is not a benchmark stunt. It is a working capacity, available across the providers Fortune 500 audit programs already approve.

For a chunk of enterprise workflows, this changes the architecture. Legal review of a 200-page contract no longer needs a chunking pipeline. M&A due diligence on a deal book can ask questions across the entire dataroom in one prompt. A staff engineer doing migration analysis can hold an entire mid-sized service in context without splitting the codebase across retrievals.

What does not change: the controls around the corpus. You still need sourcing, citations, deterministic re-runs, and an eval harness against the questions you actually care about. The model can read everything. That does not mean the model is correct, and it certainly does not mean the model output is admissible without grounding.

The teams compressing weeks of diligence into days are not switching models. They are switching their reference architecture. Long-context-first workflows look different from chunked-RAG workflows, and the 12-week migration is mostly about reviewer process, not infrastructure.