Up to 70% off kits30% off services
AI App Development Cost Calculator
What an AI feature costs to build and to run. Pick the kind of feature, enter your active users and how often they'd use it, and see the model cost per user on 9 Claude, GPT and Gemini models, what your heaviest users cost, and what share of your plan price AI would take. Prices checked 22 Nov 2026.
Your feature and users
Instructions, a few retrieved help articles and the chat so far on every message. Starting points to edit, not benchmarks.
Rule of thumb: 1,000 tokens is about 750 English words. Instructions and any retrieved help articles are sent again with every message.
Model cost on the cheapest of 9 models
$14.00 / month
Claude Haiku 5.5 · $0.01 per average user · $0.05 per heavy user
On Claude Haiku 5.5, AI costs about <0.1% of what each user pays you.
| Model | Per use | Average user | Heavy user | All users / mo | Of plan price |
|---|---|---|---|---|---|
| Claude Haiku 5.5Anthropic · SmallPrompts up to 100,000 tokens | $0.0005 | $0.01 | $0.05 | $14.00 | <0.1% |
| GPT-6 LunaOpenAI · Small | $0.0005 | $0.01 | $0.05 | $14.00 | <0.1% |
| Gemini 3.5 Flash-LiteGoogle · Small | $0.0019 | $0.04 | $0.19 | $53.20 | 0.2% |
| Gemini 3.8 FlashGoogle · MidRises to $1.50 in / $7.50 out on 1 Jan 2027 | $0.0037 | $0.07 | $0.38 | $105 | 0.4% |
| Claude Sonnet 5.5Anthropic · Mid | $0.01 | $0.20 | $1.00 | $280 | 1.0% |
| GPT-6.2 SolOpenAI · Mid | $0.01 | $0.20 | $1.00 | $280 | 1.0% |
| Claude Opus 5.5Anthropic · Large | $0.02 | $0.40 | $2.00 | $560 | 1.9% |
| GPT-6.2 AstraOpenAI · Large | $0.04 | $0.80 | $4.00 | $1,120 | 3.9% |
| Claude Fable 5.2Anthropic · Large | $0.04 | $0.80 | $4.00 | $1,120 | 3.9% |
Standard rates from Anthropic, OpenAI and Google, checked 22 Nov 2026. "All users" includes your heavy users. A highlighted row is the model used for the first-year total.
First-year cost
Your estimate. Often small; enter 0 if it runs on servers you already pay for.
- Build, one time
- from $2,400
- Running, a month
- $39.00
- First year, about
- $2,868
AI Feature Build: one ai feature in your existing web or mobile app, 3 to 4 weeks. Fixed after a $499 discovery that comes off the build.
Want measured numbers instead of estimates?30% OFF
Our $349$499 AI Feature Discovery tests your best feature idea on 30 real examples with two models and reports quality and cost per user, with a fixed build price. The fee comes off the build.
How to estimate what an AI feature costs
An AI feature has two bills: building it once, and running it every month. The running bill is mostly model usage, and it scales with how often people use the feature, not with how many sign up.
- Count uses, not users. A chat assistant might get 20 messages a month from an average user; onboarding runs once. Multiply by active users, not total accounts.
- Add up what goes in. Every message resends the instructions, the help articles it retrieved and the chat so far. That input is usually most of the tokens.
- Plan for heavy users. A small group often uses the feature many times more than everyone else. Price the plan for them, and cap usage per user.
- Compare with what users pay. If AI takes more than about 10% of the plan price for an average user, route routine messages to a smaller model, send less context, or move the feature up a plan.
- Prove it before you build. Score your ideas with the AI use case scorecard, then test the best one with the LLM evaluation template. Quality decides which model you can afford to use.
Frequently asked questions
Build cost, cost per user, heavy users, margins and where the prices come from.
With us, one AI feature added to an existing app starts at $2,400, and an AI MVP with up to 3 AI features and an admin panel starts at $5,900. Both are fixed after a $499 discovery, which comes off the build. Agencies that bill by the hour usually quote a range after their own paid discovery; ask for the test set and cost tracking to be in the quote, because they are what keeps the feature working.
Uses per user per month times the tokens each use takes times the model price. For most assistants and summaries on a small or mid-tier model that is cents per active user a month. Images, long documents and copilots that call tools cost more per use, and a small share of heavy users often drives most of the bill.
Averages hide the bill. If 10% of users use the feature five times as much as everyone else, they cost five times as much each and push the total up by about 40%. Set a per-user monthly cap and show remaining usage in the app, so a heavy user cannot run up the bill without anyone noticing.
There is no fixed rule, but if model cost for an average user is more than about 10% of what they pay you, look again before launch: route routine messages to a smaller model, send less context, cache instructions that repeat, cap heavy users, or move the feature to a higher plan.
From each provider's own pricing page (Anthropic, OpenAI and Google), standard short-context rates, checked 22 Nov 2026. Cached input, batch jobs, long prompts and image inputs are priced differently. Check the provider's page before you commit a budget.
Hosting for the feature and, for assistants that answer from your content, a vector database or search index. For most products that is a small fixed amount; enter your own estimate. Add our optional care plan if you want someone rerunning the test set and updating models each month.