FDE ForgeBeta
Start free
Week 5 · Enterprise AI

Every token has a price

0:00 / 3:00

20 seconds

The demo burned $40.

Helix Health's pilot answered 200 questions with a frontier model and no ceiling. Finance noticed before clinical ops did.

Question2k tokens inModel$ / 1M tokensCostin + out
Lesson notesEvery token has a price: the short version+

Week 5 · LLM gateways · about 3 minutes

What you will be able to do

Estimate cost per Knowledge Copilot request before the Helix Health demo.

The idea

Every request costs (input_tokens × in_price + output_tokens × out_price) / 1_000_000. Context from RAG multiplies the input side.

Before a demo, multiply expected questions by a high-end estimate. If the number scares you, shrink context or pick a smaller model first.

def request_cost(in_tokens, out_tokens, in_per_m, out_per_m):
    return (in_tokens * in_per_m + out_tokens * out_per_m) / 1_000_000

# 2k in, 400 out at $5 / $15 per 1M → $0.016 per answer
assert round(request_cost(2000, 400, 5.0, 15.0), 4) == 0.016

If you cannot estimate the demo bill, you are not ready to demo.

Key takeaways

  • Cost = (in×in_price + out×out_price) / 1M.
  • RAG context dominates the input bill.
  • Estimate the demo before you run it.

Part of the Forward Deployed Engineer roadmap. New to the role? Start with what a Forward Deployed Engineer does.