Calculator
What do runaway loops and power users cost you?
Four inputs, list prices for 30+ models, and every formula printed under its result. Change any assumption and the numbers recompute.
Assumptions you can change
- Monthly margin leak from power users
- $6,507
- 20 power users drive $7,200 (40% of $18,000); budgeted at $34.65 each = $693. Leak = $7,200 - $693.
- Annual runaway loop exposure
- $18,123
- 1 incident/month x 12 x $1,510 per uncaught loop (2,500 calls in 45 min). With breakers: $23.00 a year.
- Monthly savings from cache and fallback
- $8,917
- Cache: $18,000 x 85% input share x 30% hit rate x 90% cache discount = $4,145. Fallback to Claude 3.5 Haiku recovers 73% of the leak = $4,772.
- Cost of a normal session
- $1.44
- 12 steps on Claude Sonnet 4.5, starting at 24,000 input tokens and growing 1,800 per step.
Recommended plan
PALM SOUQ Growth$499/month
Annual benefit $125,105 (loop exposure avoided plus 12 months of cache and fallback savings) against $5,988 a year: 21x return.
Method
How each number is calculated.
Margin leak
power-user spend - (power users x per-tenant token budget)
Power users are the top share of tenants by token spend. Their budget is a share of what they pay you. Anything above it is margin you give away on a flat-rate plan.
Loop exposure
incidents/month x 12 x cost of 2,500 calls
An uncaught loop is modelled as 2,500 calls over 45 minutes, the shape of a coding agent stuck on a merge conflict. With breakers on, the same session ends at call 15.
Session cost
sum over steps of (input_i x input price + output x output price)
Input tokens start at the first-call size and grow by the context added per step, capped at the model's context window. That is why long loops get more expensive per call.
Cache savings
invoice x input share x hit rate x (1 - cached price / input price)
Uses the cached-input price for the selected model. Models with no cache pricing return zero.
Fallback savings
leak x (1 - fallback price / primary price)
Instead of cutting power users off at their cap, traffic above it moves to the cheapest model from the same provider. The blended price ratio uses your input and output mix.
Return
(loop exposure avoided + 12 x monthly savings) / annual plan cost
Plan recommendation follows declared spend and tenant count: Starter under $5k a month and 50 tenants, Growth to $30k and 1,000 tenants, Scale above.
Worked example
A support agent SaaS with 500 customers.
500 customers at $99 a month is $49,500 MRR. 25 power users burn $400 a month each in raw tokens, so each of them costs you $301 more than they pay.
Capping their token spend at $35 with a per-tenant quota, and offering usage overages above it, recovers the difference every month.
power users 25
raw spend each $400/month
quota cap each $35/month
recovered = 25 x ($400 - $35)
= 25 x $365
= $9,125/monthCalculator questions
Where do the model prices come from?
From the PALM SOUQ LLM Price Index, reviewed June 2026. The calculator, the index and the gateway sandbox all read the same price table, so a price correction shows up everywhere at once.
Why does a runaway loop cost more per call over time?
Agents append each step's output and tool results to the conversation. A loop that keeps retrying sends a longer prompt every time until it hits the model's context window, so the cost curve bends upward before it flattens.
How many loop incidents should I assume?
Use your own incident history. If you have never measured it, start with one a month and check the breaker events in your first week of the PALM SOUQ trial; the event log gives you the real number.
Does the report include my prompts or customer data?
No. The report only uses the numbers you entered. PALM SOUQ never captures prompt text, completion text or tenant personal data in analytics, and the gateway itself runs with Zero Payload Retention.
What does the emailed report contain?
Your inputs, every formula with the values filled in, the recommended plan, and a starting breaker and quota policy per tier that you can load into the gateway during your trial.
Want to see the breaker stop a loop in real time? Open the gateway sandbox or compare model prices in the LLM Price Index.
Turn these estimates into enforced limits.
Route your agent traffic through PALM SOUQ in 2 minutes. Change one base URL, pass a tenant header, and every token is attributed and governed.
14-day free production trial · 100k requests included · no card required