Deep Research Cost per Seat

Deep research products are usually sold like seats, but the real cost driver is the number of long, retrieval-heavy sessions each seat runs every month.

Question

How do I estimate deep research cost per seat?

Quick answer

Formula: cost_per_seat_month = sessions_per_seat_month * cost_per_session

Assumption: a session bundles the full research interaction, not one prompt-response pair.
Assumption: premium-model output and wide retrieval can dominate session cost fast.
Assumption: architecture comparisons keep the quality bar constant across RAG and long-context options.

Example: 35 sessions per seat-month at $0.72 per session gives $25.20 per seat-month.

When Per-session Cost Spikes

Retrieved context grows wide because the task pulls from many sources.
Output length expands into long synthesis or structured reporting.
Premium models stay on the hot path for most of the session.
Long-context prompts replace retrieval without actually reducing token load enough.

Architecture Check Before Pricing

If a long-context design removes enough retrieval, reranking, vector, and refresh overhead, it can beat a RAG stack even when the prompt is larger. If it does not, the larger prompt simply layers extra token spend on top of an already expensive workflow.

Deep research should be priced only after that architecture comparison is clear, because a small change in session cost compounds hard when each seat runs long sessions all month.

Recommended Next Step

Explore provider options for a deep research workflow after you've compared long-context and retrieval-heavy designs.

View Infra Recommendations