# Pricing an unlimited AI plan against a tail you haven't measured

> Pricing an unlimited AI plan means pricing against a maximum nobody has measured. What unlimited really costs, and the safer alternatives.

URL: https://getculpa.com/price-an-unlimited-ai-plan
Last reviewed: 2026-08-01

## Answer
An unlimited AI plan prices a promise against a maximum nobody has computed, and on a long-tailed usage curve that maximum keeps moving. A generous included allowance with a stated overage reads nearly as well and bounds the risk. Culpa, a local-first LLM cost, margin, and forecast ledger, reports the spend distribution so the tail is a number rather than a fear.

## Why this happens
Unlimited works in traditional SaaS because marginal cost is near zero. On an AI product marginal cost is the dominant cost, so unlimited transfers the entire variance of your usage curve onto your own margin. The damage is concentrated: a handful of accounts discover the ceiling doesn't exist and build workflows around that fact. Those accounts are usually your most engaged, which makes the conversation afterwards harder, not easier.

## What this usually looks like
- Your unlimited tier has accounts costing multiples of what they pay.
- Nobody has computed what the heaviest plausible account would cost on the plan.
- The unlimited plan was priced against competitor positioning rather than your own unit costs.
- Usage on that tier grows faster than on metered tiers, which is the plan working as designed against you.

## Common mistakes
- Launching unlimited before measuring your own tail. Why it hurts: You've priced a promise against a maximum that doesn't exist yet and will keep rising. Do instead: Measure the 99th percentile first, price against that, and only then decide whether unlimited is affordable.
- Matching a competitor's unlimited plan. Why it hurts: Their cost structure, model mix and usage curve aren't yours, so the same price means a different margin. Do instead: Price from your own measured distribution. A plan that works for them can be fatal for you.
- Treating fair-use language as a cost control. Why it hurts: Fair-use clauses are rarely enforced and never enforced early, so the spend lands before the conversation. Do instead: Implement an actual ceiling with alerting, then handle exceptions commercially rather than in the terms.

## Self-check
- Compute what your single heaviest account costs today, then double it. Can the plan absorb that?
- Find your 99th percentile account cost, not the 90th. Unlimited is priced against the extreme.
- Check whether usage growth on the unlimited tier outpaces your metered tiers.
- Ask what happens if your three heaviest accounts all double next quarter.
- Confirm you'd know within days rather than at month end.

## What unlimited costs when the tail moves (illustrative)
An unlimited tier at $200 a month on Claude Haiku 4.5 for volume work at $1.00 per million input and $5.00 output. Usage figures are modelled.
  Typical account: 20M input, 4M output = (20 x $1.00) + (4 x $5.00) = $20 + $20 = $40, margin 80%
  Heavy account: 150M input, 30M output = $150 + $150 = $300, margin negative 50%
  Extreme account: 400M input, 90M output = $400 + $450 = $850, margin negative 325%
  Ten typical accounts fund one heavy account, and can't fund one extreme account at all
  At 5% extreme accounts the tier loses money regardless of how many typical accounts you add
Growth stops rescuing an unlimited tier once the extreme share crosses a few percent, because each new typical account adds $160 of margin and each extreme one removes $650.

## Cost figures
Every figure carries its confidence and its source. No figure on this site is provider-reported.
- $40 to $850 — modelled monthly cost range on a single $200 unlimited tier [estimated] Source: Both endpoints at real Claude Haiku 4.5 rates per 1k tokens from the price book, effective 2026-07-02. A range because the usage profiles are modelled.
- -325% to 80% — modelled gross margin spread from typical to extreme account on a $200 unlimited tier [estimated] Source: ($200 - $850) / $200, from the teardown arithmetic at real Claude Haiku 4.5 rates. Modelled usage, so published as a scenario.

## FAQ
Q: Is an unlimited AI plan ever a good idea?
A: It can work where your marginal cost per user is genuinely small and bounded, or where the plan is a deliberate loss leader you've sized. It fails when it's chosen to match a competitor without measuring your own tail first.

Q: What's the safer alternative to unlimited?
A: A generous included allowance with a stated overage rate. It reads nearly as well in marketing, it bounds your risk, and it gives heavy users a clear signal before they become a problem rather than after.

Q: Does fair-use language protect me?
A: In practice, rarely. Fair-use clauses are enforced late if at all, so the spend has already landed by the time anyone invokes them. An actual ceiling with alerting does the job the clause is imagined to do.

## Sources
- Anthropic pricing: https://platform.claude.com/docs/en/docs/about-claude/pricing

Run the free Cost Leak Scan: https://app.getculpa.com/scan?source=pseo&slug=price-an-unlimited-ai-plan&cluster=margin

Machine-readable index of every guide: https://getculpa.com/api/pages
Human-readable index of every guide: https://getculpa.com/guides
Site overview: https://app.getculpa.com/llms.txt

Privacy: Culpa runs on your infrastructure. Your prompts and responses never leave it. Culpa counts calls to run your plan, and it fails open, so if it ever breaks your app keeps running.