Pricing, Explained

Why we show you this

Most software companies hide their costs. They don't want you to know the margin, or how much the AI APIs actually cost, or how much of your subscription goes to cloud infrastructure versus the vendor's pocket. Opacity is the default, and it is usually intentional.

We do the opposite. This page shows, line by line, what it costs to run OmniThink — the AI API bills, the infrastructure, the payment fees — and exactly how the prices are derived from those costs. We do it because we think you deserve to know, and because the only way to ask for your support honestly is to first show you what you are supporting.

What one question costs us

When you ask a question, it is sent to all five models at the same moment. Every token each model reads and writes is billed to us at that provider's published API rate. A typical question is about 800 tokens in (your question plus any active Answer Modes) and 600 tokens out per model. At the current rates (verified July 2026):

Running Compare or Aggregate afterwards is a sixth call — Claude reads all five answers (≈4,800 tokens in, 800 out) and synthesises them: about $0.044 more. So a full session — five models plus a Compare — costs us roughly 10 cents in raw AI fees. Answer Modes lengthen the input; multi-round Debate re-sends the growing transcript each round and can cost several times more per question.

What it costs to keep the lights on

Behind every question there is production infrastructure running continuously, whether anyone is using it or not:

Plus payment processing (Stripe keeps 2.9% + $0.30 of every charge: 53¢ of a Plus payment, 73¢ of a Pro payment) and tax on revenue.

The plans, and how the prices were derived

First, the unit. Since July 2026 we meter the fairest way we know: one answer = one model's reply, and you are charged exactly the answers you request. Ask all five models and that's 5 answers; ask one and it's 1 — lighter questions stretch your plan further. A Compare or Meta-answer costs 1; a Debate costs one per model per round plus the conclusion. (When another service advertises thousands of "messages," check whether they mean one reply from one lightweight model. Every Omnithink answer is a frontier model.)

The derivation is straightforward, and now it closes. Take Pro: of $14.99, Stripe takes $0.73 and we provision 15% for tax, leaving about $12. A subscriber who burns the entire 800-answer quota costs about $9 in AI fees. Plans are sized to a published rule — even at 100% utilization, at least a 20% margin must remain — because the earlier versions of this page had to admit that a maxed-out subscriber cost more than they paid, with the founder absorbing the difference. That is no longer true: the margin funds the infrastructure above and the free tiers below, and there is still no hidden multiple on top — the whole formula is on this page.

The same holds for Plus at its 400-answer scale (about a 22% margin at full burn). Free tiers are funded by the paid ones: a free account costs us at most about a dollar a month, and a guest a few cents a day — that is the marketing budget, spent on you trying the product instead of on ads.

Past your plan there's no wall: Omnis, our prepaid coins, cover overflow. 100 Omnis = $1; an answer burns its real AI cost × 4, stated exactly that plainly in the app. They never expire, you can earn them free (signing up, referrals, promos), and subscriptions remain the better per-answer deal — Omnis price flexibility, not loyalty.

Compared with subscribing directly

The catch, stated honestly: those direct subscriptions include unlimited-ish chat with one model. OmniThink gives you a monthly budget of questions answered by all five at once, with Compare, Debate, and 46 Answer Modes on top. Different shape, radically lower price for what it does.

Fair use, and what keeps this sustainable

• One quota pool. Your plan's monthly quota is shared between the OmniThink app and partner apps that use your account — one subscription, one pool.
• Personal use only. No automation, scripting, or resale. Quotas and rate limits enforce this technically, not just contractually.
• Bounded answers. Model responses are capped at generous lengths so a single request can't consume unbounded compute.
• We eat price changes first. When providers change API prices (OpenAI doubled theirs in April 2026), we update this page and absorb the difference until a considered re-pricing — we don't silently degrade the product.

The bottom line

The math is honest. The price is real. If you use OmniThink regularly and find it valuable, subscribing is the most direct way to say: this should continue to exist.

See the plans →