Long Context Drives Vendor Lock-In

Long context becomes sticky when it stops being a model feature and becomes part of the customer’s operating workflow. Teams that analyze contracts, filings, manuals, or research packets start designing prompts, templates, review steps, and internal tools around feeding whole documents into one pass, instead of chunking, retrieval tuning, or manual copy and paste. Once that workflow is wired in, switching vendors means rebuilding both the technical pipeline and the human process around it.

• Moonshot reinforced this with context caching. Its developer docs show caching for repeated large prompts can cut cost by up to 90% and improve latency, which makes always on document workflows cheaper to run and easier to productize into apps, copilots, and internal QA tools.
  1 moonshot 2 moonshot
• The closest analogue is Anthropic. Long context and prompt caching helped make Claude easier for developers doing document RAG, multi shot prompting, and codebase analysis. That shows why context depth can drive adoption even when raw model quality is similar across vendors.
  5 sacra
• The limit is that long context alone does not fully lock customers in. In production, many teams still fine tune or add a separate workflow layer because very large prompts can be slow and expensive, and vertical products increasingly win by owning templates, approvals, evaluations, and document specific actions rather than the base model alone.
  3 sacra 4 sacra

This is heading toward a split market. Base models will keep matching each other on context length, which weakens context as a standalone moat. The durable winners will be the companies that turn long context into a full document operating system, with caching, workflow logic, and accumulated organization specific context that customers do not want to rebuild elsewhere.