---
title: "LLM cost per user: SaaS AI feature unit economics | BroadRouter"
description: "Work out what an AI feature costs per active user per month, why heavy users matter more than averages, and how to price SaaS tiers so the model bill stays covered."
url: https://broadrouter.com/blog/llm-cost-per-user-saas/
lang: en
updated: 2026-11-14
publisher: BroadRouter (a BroadNet site); API by BroadNet Inference, https://inference.broadnet.ai
---

# LLM cost per user: unit economics for a SaaS AI feature

By the BroadNet Inference team · 6 min read · Published 2026-11-14

LLM cost per user = requests per active user per month × tokens per request × the per-1M token prices. Measure it per feature from the usage field, price tiers on the 90th-percentile user rather than the average, and add fair-use limits for the tail. Cheaper models, caching and low reasoning effort cut the number directly.

## Key takeaways

-   Cost per user = requests per user × (input, cached and output tokens per request) × per-1M prices.
-   Measure by feature and by user from the usage field; averages hide the heavy users who drive the bill.
-   Model choice moves cost per user several-fold: compare output prices such as $0.22, $0.52 and $3.46 per 1M.
-   Price tiers on your 90th-percentile user and add fair-use limits for the tail.
-   Caching, reasoning\_effort and routing cut cost per user without changing the product.

**On this page**

1.  How do you calculate LLM cost per user?
2.  Step 1: What counts as an active user, and what is the unit?
3.  Step 2: How do you measure tokens per unit?
4.  Step 3: What does a typical user cost? A worked example
5.  Step 4: Why does the average user mislead?
6.  Step 5: How do you set tiers and limits?
7.  Step 6: How do you cut cost per user without changing the product?
8.  Can a flat-rate plan serve SaaS users?
9.  What costs sit around the model?
10.  How should you handle free tiers and trials?
11.  Summary

Adding an AI feature to a SaaS product changes its cost structure. Hosting and storage cost roughly the same for every user; LLM calls cost more the more someone uses them. If you price the feature without knowing what an active user costs you, a few enthusiastic customers can turn a profitable plan into a loss. This post shows how to compute LLM cost per user, why the average is the wrong number, and how to use it to set tiers.

## How do you calculate LLM cost per user?

For one feature and one model:

```text
cost per user per month =
    requests per active user per month
  × [ new input tokens × input price
    + cached tokens    × cached price
    + output tokens    × output price ] / 1,000,000
```

Output includes reasoning tokens on reasoning models, and images are counted as input tokens on a pay-as-you-go key. If you need a refresher on those three prices, read [how LLM API pricing works](https://broadrouter.com/blog/llm-api-pricing-explained/).

Add up the features a user touches to get the total AI cost per user.

## Step 1: What counts as an active user, and what is the unit?

Pick a definition you already report on, such as a user who used the AI feature at least once in the month. Then pick the unit your feature naturally produces: one chat turn, one generated document, one analysed file, one summarised call. Cost per unit is more stable than cost per user and easier to reason about.

## Step 2: How do you measure tokens per unit?

Do not estimate this from word counts. Log the `usage` block of every response with a feature tag and a hashed user or account ID: prompt tokens, cached tokens and completion tokens, one row per attempt. The [usage logging post](https://broadrouter.com/blog/llm-api-usage-logging/) has a ready wrapper and the SQL to group cost by user and feature. Before you have traffic, the [token counter](https://broadrouter.com/tools/token-counter/) gives a first estimate from sample prompts.

For streamed replies, request usage in the last chunk with `stream_options={"include_usage": True}`; see the [streaming guide](https://broadrouter.com/guides/streaming-chat-completions/). Count every attempt, including retries, because every attempt that reaches the model is billed.

## Step 3: What does a typical user cost? A worked example

Suppose your product has an in-app assistant. You measure, for a typical active user in a month:

| Measure | Value |
| --- | --- |
| Active days | 20 |
| Assistant requests per active day | 15 |
| Input tokens per request | 3,000 (half cached) |
| Visible answer tokens per request | 400 |
| Reasoning tokens per request (reasoning models) | about 600 |

That is 300 requests a month: about 900,000 input tokens, 450,000 of them cached, and 120,000 visible output tokens, plus about 180,000 reasoning tokens on a reasoning model.

Apply BroadNet’s per-1M prices. Token counts are in millions, so each term is millions of tokens × price per 1M:

| Model | Input / cached / output per 1M | This user’s month |
| --- | --- | --- |
| [Gemma 4 VL 26B](https://broadrouter.com/models/gemma-4-vl-26b/) | $0.10 / no cached rate / $0.22 | 0.9 × $0.10 (all input at full price) + 0.12 × $0.22 (no reasoning tokens) = **$0.12** |
| [DeepSeek V4.1 Flash](https://broadrouter.com/models/deepseek-v4.1-flash/) | $0.175 / $0.0055 / $0.52 | 0.45 × $0.175 + 0.45 × $0.0055 + 0.3 × $0.52 = **$0.24** |
| [Qwen3.8 27B](https://broadrouter.com/models/qwen3.8-27b/) | $0.175 / $0.0435 / $2.17 | 0.45 × $0.175 + 0.45 × $0.0435 + 0.3 × $2.17 = **$0.75** |
| [GLM 5.3](https://broadrouter.com/models/glm-5.3/) | $0.78 / $0.173 / $3.46 | 0.45 × $0.78 + 0.45 × $0.173 + 0.3 × $3.46 = **$1.47** |

The 0.3M of output on the reasoning models is 0.12M of visible answer plus 0.18M of reasoning. Worked through, the GLM 5.3 line comes to more than ten times the Gemma 4 VL 26B line for the same user (the [Gemma 4 VL 26B vs GLM 5.3 comparison](https://broadrouter.com/compare/gemma-4-vl-26b-vs-glm-5.3/) shows the two side by side). Output is the largest term on every reasoning model, and reasoning tokens are more than half of that output. Put your own measurements into the [LLM API cost calculator](https://broadrouter.com/tools/llm-api-cost-calculator/) to get your figure.

## Step 4: Why does the average user mislead?

Usage of AI features is skewed. Most users try the feature a few times; a small group uses it all day. The average hides that group, and that group drives the bill.

From your usage log, compute per user per month:

-   **Median cost.** What a normal user costs.
-   **90th percentile.** What a heavy but normal user costs. Price your tier so this user is still profitable.
-   **99th percentile and maximum.** The tail. These users need a fair-use limit, a higher tier, or a check for automation and abuse.

If your 99th-percentile user costs many times your median, that is not a problem by itself; it is a pricing input.

## Step 5: How do you set tiers and limits?

With cost per user known, pricing becomes a margin decision:

1.  **Decide your target gross margin** for the tier.
2.  **Work back to the maximum AI cost per user** that margin allows.
3.  **Compare it with your 90th percentile.** If the 90th percentile fits, the tier works. If not, reduce cost (Step 6), raise the price, or move heavy use to a higher tier.
4.  **Translate tokens into a unit users understand.** Users do not think in tokens. Limit requests per day, documents per month or minutes of audio, sized so the heaviest allowed use stays inside your cost ceiling.
5.  **Re-check monthly.** Prompt changes, new features and model changes all move cost per user.

## Step 6: How do you cut cost per user without changing the product?

-   **Cache.** Put the stable system prompt and tool list first. On DeepSeek V4.1 Flash, cached input costs $0.0055 per 1M instead of $0.175. [How to raise your cache rate](https://broadrouter.com/blog/cached-input-tokens-explained/).
-   **Turn reasoning down.** `reasoning_effort` set to `low` for routine requests cuts the most expensive tokens. [What reasoning costs](https://broadrouter.com/blog/reasoning-tokens-cost/).
-   **Route by difficulty.** Send simple requests to Gemma 4 VL 26B and keep the reasoning models for hard ones.
-   **Trim history.** Long conversations resend everything each turn; summarise or truncate older turns where quality allows.
-   **Cap output.** Set `max_tokens` per feature, leaving room for reasoning on reasoning models.

More controls are in [LLM API spend control](https://broadrouter.com/blog/llm-api-spend-control/).

## Can a flat-rate plan serve SaaS users?

BroadNet’s unlimited plans for DeepSeek V4.1 Flash ($60 a month per session) may be built into your own app or service under your own key, but each session runs **one request at a time** and the key cannot be shared or resold. That makes a plan a good fit for back-end work with low concurrency: a nightly report generator, a queue-driven summariser, an internal agent. For interactive user traffic with many simultaneous requests, pay-as-you-go keys (4 requests in flight each, 30 per minute) or several sessions are the usual answer. Check the [plans page](https://inference.broadnet.ai/plans?utm_source=broadrouter&utm_medium=referral&utm_campaign=blog-llm-cost-per-user-saas&utm_content=body) for current terms.

## What costs sit around the model?

Cost per user from tokens is only part of the AI line. Add retrieval and vector storage, transcription (priced per audio minute, for example $0.0042 per minute for Arabic on BroadNet), retries and evaluation runs. If your users write in Arabic or other non-English languages, expect more tokens per message than in English; [measure tokens per word](https://broadrouter.com/blog/tokens-per-word-arabic-english/) for your own content.

## How should you handle free tiers and trials?

A free tier multiplies cost per user by users who pay nothing. Treat it as marketing spend: give free users the cheapest model that still shows the feature well, set tighter daily limits than on paid tiers, and track free-user AI cost as its own line. If free users convert, that line is acquisition cost; if they do not, the limits keep it small.

## Summary

Measure tokens per unit from the usage field, multiply by published prices, look at the 90th percentile rather than the mean, and set tier prices and limits from that. Then make every unit cheaper with caching, routing and lower reasoning effort. Start with the [monthly cost method](https://broadrouter.com/blog/estimate-llm-api-cost-before-launch/) if you have not measured anything yet.

Free tools for this

-   [LLM API cost calculator](https://broadrouter.com/tools/llm-api-cost-calculator/) Price a workload on every model at once, with cache hits, reasoning and a flat-plan check.
-   [Flat-rate vs per-token LLM cost calculator](https://broadrouter.com/tools/flat-rate-vs-per-token-calculator/) Break-even between an unlimited plan and paying per token, per agent.
-   [Prompt caching savings calculator](https://broadrouter.com/tools/prompt-caching-calculator/) Bill with and without caching, by prefix share and hit rate.
-   [Prompt token diff](https://broadrouter.com/tools/prompt-token-diff/) Two prompt versions side by side: changed lines, exact token delta and the monthly cost difference.

[All free tools](https://broadrouter.com/tools/)

## Terms in this article

[Token](https://broadrouter.com/glossary/token/)

A token is the unit of text a language model reads and writes: a word, part of a word, a punctuation mark or a byte sequence produced by the model's tokenizer. LLM APIs count, limit and usually bill usage in tokens.

[Input and output tokens](https://broadrouter.com/glossary/input-and-output-tokens/)

Input tokens are the tokens a language model reads in a request, including the system prompt, conversation history, tool definitions and images. Output tokens are the tokens it generates in reply, including any reasoning. LLM APIs usually price them separately, and output usually costs more.

[Reasoning tokens](https://broadrouter.com/glossary/reasoning-tokens/)

Reasoning tokens are tokens a reasoning model generates to work through a problem before it writes its visible answer. They are usually billed as output tokens, count toward the output limit and the context window, and may be hidden or summarized by the API.

[Pay-as-you-go](https://broadrouter.com/glossary/pay-as-you-go/)

Pay-as-you-go is a pricing model in which you pay only for what you use, metered per unit such as tokens, requests or audio minutes, with no subscription or minimum commitment. Most LLM APIs bill this way, from a prepaid balance or a periodic invoice.

[Concurrency](https://broadrouter.com/glossary/concurrency/)

Concurrency is the number of requests a client has in flight at the same time. LLM APIs often cap it per key or plan, separately from requests per minute, because each open request holds GPU capacity until its response has finished.

[Tool calling](https://broadrouter.com/glossary/tool-calling/)

Tool calling, also called function calling, is an LLM API feature in which the model returns a structured request to run a function the developer has defined, with JSON arguments. The application runs the function and sends the result back; the model never runs code itself.

[Cached input tokens](https://broadrouter.com/glossary/cached-input-tokens/)

Cached input tokens are prompt tokens that an LLM provider has recently processed and can reuse from a cache instead of computing again, usually a repeated prefix such as a system prompt or document. Providers bill them at a lower rate than new input tokens.

[Rate limit](https://broadrouter.com/glossary/rate-limit/)

A rate limit is a cap an API places on how much a client can use it in a time window, such as requests per minute or tokens per minute. Requests over the limit are refused, usually with HTTP 429 Too Many Requests, until usage falls back under it.

## Questions

### Should I charge users per token?

Rarely for end users; most prefer a flat tier with a clear limit. Measure tokens internally, price in requests, documents or seats, and keep a fair-use limit for the heaviest tail.

### Can I use a flat BroadNet plan to serve my SaaS users?

BroadNet allows building a plan into your own app or service under your own key. Each session runs one request at a time, so it suits back-end agents or low-concurrency features; user-facing traffic with many simultaneous requests usually needs pay-as-you-go keys or several sessions.

### What share of a subscription price should the AI cost be?

There is no universal number. Decide the gross margin your business needs, then work backwards to the most a heavy normal user may cost and set limits to match.

### How often should I recompute cost per user?

Whenever you change the prompt, the model, or the feature, and at least monthly. Token counts drift as prompts and user behaviour change.

## Keep reading

-   [How to estimate LLM API cost before you launch a feature](https://broadrouter.com/blog/estimate-llm-api-cost-before-launch/)
-   [LLM API spend control: budgets, key limits and loop protection](https://broadrouter.com/blog/llm-api-spend-control/)
-   [Reasoning tokens: why you pay for more output than you see](https://broadrouter.com/blog/reasoning-tokens-cost/)
-   [Cached input tokens: how to cut an LLM API bill](https://broadrouter.com/blog/cached-input-tokens-explained/)
-   [LLM API price changes in October 2026: 3 rises, no cuts](https://broadrouter.com/blog/llm-api-price-changes-october-2026/)
-   [Do LLM API prices only go down? 52 rises and how to budget for them](https://broadrouter.com/blog/llm-api-price-increases/)
-   [Guide: Use the OpenAI Python SDK with BroadNet Inference](https://broadrouter.com/guides/openai-python-sdk/)
-   [Guide: Stream chat completions and handle dropped connections](https://broadrouter.com/guides/streaming-chat-completions/)

Try it on your real workload. New accounts start with $5 of credit for Gemma and transcription. DeepSeek, GLM and Qwen3.8 need a top-up.

[Get started](https://inference.broadnet.ai/account/?utm_source=broadrouter&utm_medium=referral&utm_campaign=blog-llm-cost-per-user-saas&utm_content=page-cta) ·

**Sources**

-   [BroadNet Inference machine-readable prices (pricing.json)](https://inference.broadnet.ai/pricing.json?utm_source=broadrouter&utm_medium=referral&utm_campaign=blog-llm-cost-per-user-saas&utm_content=sources)
-   [BroadNet Inference API reference, usage fields and limits](https://inference.broadnet.ai/docs?utm_source=broadrouter&utm_medium=referral&utm_campaign=blog-llm-cost-per-user-saas&utm_content=sources)
-   [BroadNet Inference unlimited plans and plan FAQ](https://inference.broadnet.ai/plans?utm_source=broadrouter&utm_medium=referral&utm_campaign=blog-llm-cost-per-user-saas&utm_content=sources)
-   [OpenAI API reference, chat completions usage object](https://platform.openai.com/docs/api-reference/chat/create)
