Titan
TitanAI Cost Simulator
What it costs to run TitanAI per user — and what you'd make at a given price
How much does a user use?
Usage per user / month ?A user's monthly AI usage in TitanAI tokens — the underlying billable unit. See "Assumptions & technical detail" for what a TitanAI token covers and how the rate is derived.
13.5M
TitanAI tokens / month
Number of users ?How many paying users you're serving. More users spread the fixed costs thinner.
If you charge a markup of 100%
A 100% markup means you charge 2× what it costs — a 50% profit margin.
You charge / user / mo
$22.50
You keep / user / mo
$11.25
50% margin
Revenue / month $11,250
Profit / month $5,624
Profit / year $67,491
What it costs us
$11.25
per user / month
All users / month
$5,626
All users / year
$67,506
Where it comes from · per user / month
AI usage ?The model API cost — your monthly tokens × the blended token rate. The rate is built from the model mix; see "Assumptions & technical detail" for the derivation and the underlying provider.$10.94
13.5M TitanAI tokens × $2.17 / 1M
Infrastructure ?Servers (scale with usage) + Database, Web hosting, Observability (currently flat — today's invoice snapshot, not auto-scaling). Bump the components in "Assumptions & technical detail" at much larger scale.$0.32
Servers (1 × $60)$60/mo
Database$18/mo
Web hosting$48/mo
Observability$32/mo
$158/mo ÷ 500 users
Cost per user / month$11.25
× 500 users$5,626 / mo
What's a TitanAI token? One token the agent processes end-to-end — your message plus the ~13K-token system prompt, tool calls, and multi-agent fan-out, spread across ~9.65 OpenAI calls per prompt. So it's billed via OpenAI, but it's far more than the text a user types.

Where the rate comes from: the architecture is gpt-5.5 orchestrator + gpt-5.4-mini workers. We price the TitanAI token as a blend of those two — ~1/3 gpt-5.5 ($5/1M, list) + ~2/3 gpt-5.4-mini ($0.75/1M, list) = ~$2.17/1M tokens. Model prices are in pricing.ts.

Why this is pessimistic vs the eval bill. The May 2026 test billed lower ($176.91 / 217.4M ≈ $0.81/1M) because OpenAI auto prefix-caches the repeated ~13K system prompt across the ~9.65 calls per prompt. We deliberately price off list rates (no caching benefit baked in) so pricing doesn't break if that automatic caching shifts or production usage changes.
AI Token Cost
$
$
%
Blended rate = share × gpt-5.5 + (1−share) × mini = $2.17 / 1M. Pessimistic on purpose: input list prices, no caching baked in (caching is a separate toggle below).

Two things this rate doesn't model separately: (1) Output tokens cost ~6× input on gpt-5.5 ($30/1M vs $5). For long-response usage, nudge the rates up — current rate assumes input-dominant volume. (2) Rates are list as of 2026-05-13 per pricing.ts; if the provider re-prices, edit here.
Caching
%
%
Discounts repeated tokens up to ~90%. No production data yet — default is the measured no-cache rate.
Infrastructure / mo
$
$
$
$
M tok/mo
Database / Web / Observability are the current invoice — they don't auto-scale here. At much larger scale they'd grow (more storage, queries, ingest), so bump them manually for those scenarios. Servers scale automatically on token throughput, but multi-server orchestration hasn't been built in production yet (one server today).
Other Costs
$/user
Not in the technical bill. Add it to model the full cost-to-serve, not just API + infra.
Heavy-User Risk
A 250-prompt power user (≈67.5M tokens) costs ~$55/mo vs the ~$11 average. A flat price across uneven usage eats into margin.
Live Readout
Effective rate$0.81 / 1M
Extra-usage price$1.62 / 1M
Servers needed1
Total tokens / mo6.8B