Free tool
AI Feature Cost Calculator
Estimate the monthly operating cost of an AI feature before you build it. Pick a pattern, adjust the assumptions, and see where the money goes — inference, embeddings, database, compute — plus margin and breakeven at your chosen price.
Start from a preset
Cohort-segmented digests for internal teams, customer communities, or member networks.
How often each user triggers AI work
Context + prompt size
Avg response length
Multi-step agents > 1
Real-world is typically 1.2–1.8×
Prices shown per 1M input / output tokens (USD).
What you charge — sets margin and breakeven
Projected at 0.1×, 0.5×, 1×, 2×, 5×, 10× current users.
* Model pricing as of June 15, 2026. Subject to provider change — the calculator pulls weekly updates from OpenRouter via a GitHub Action.
These are baseline operating costs
Real projects typically land 20–40% above baseline once retry rates, caching strategy, multi-region overhead, and observability are accounted for.
Architecture choices can shift the answer 5–10× — for example, a morning-locked digest versus a live-updating one, or batching versus per-request inference. A calculator can't make those calls for you.
How to read these numbers
The calculator multiplies tokens by current provider prices, adds embedding and infrastructure costs, and divides by users. That's the easy 60% of the question.
The other 40% — what most teams miss — is in the gap between a clean happy path and what actually runs in production. Retries triggered by transient errors. Long-tail conversations that blow past token budgets. Multi-region replication. Compliance logging. Cache warmup costs at deploy time. The architecture you choose can move the real number by an order of magnitude.
Use this tool to get your feet under you. Use a fractional CTO to get the answer right.
Frequently asked
Does this calculator include retry and error overhead?+
Only via the retry-factor input, which is a flat multiplier. Real systems have variable retry behaviour driven by model error rates, network conditions, rate limits, and how aggressive your client-side timeouts are. Most production AI apps run at a 1.2–1.8× effective request multiplier once you account for those — but the actual shape needs telemetry, not estimation.
How accurate is this for my specific stack?+
Treat the output as a baseline, not a forecast. It assumes a clean happy path — no caching layer, no streaming, no batching, no multi-region replication, no compliance overhead. Architecture decisions can move the real number by 5–10× in either direction. The calculator gets you in the right order of magnitude; getting the actual number right is a modelling exercise on your specific stack.
What isn't modelled here?+
Development cost, retry budgets beyond a flat factor, caching strategy savings, observability and logging infrastructure, multi-region replication, egress bandwidth, compliance and audit overhead, vector database costs separate from your primary DB, fine-tuning, RLHF, evaluation infrastructure. These are non-trivial line items at scale.
Why do prices vary so much between models?+
Frontier models price for reasoning quality and context window. Small models price for throughput. The right choice is rarely "the cheapest" — most production apps route easy queries to a small model and hard queries to a frontier model, which the calculator deliberately doesn't model. Cascading is one of the highest-leverage cost decisions, and it depends entirely on what kind of work you're asking the model to do.
How fresh are the prices?+
Model pricing updates weekly via a GitHub Action that pulls from OpenRouter's public model registry. The "pricing as of" date under the results panel shows the last refresh. If you spot a stale entry, drop us a note — the data lives in a public, version-controlled JSON file.
What happens if I email the summary?+
You get a one-page HTML email with the scenario you ran. No newsletter, no autoresponder sequence. We may follow up once if your scenario looks like something we can help with — that's it.
Want a custom model for your stack?
We build cost forecasts grounded in your real architecture — retries, caching, infra choices, and dev cost included. Book 30 minutes and we'll walk through your numbers.