AI Cost Calculator

Identify the deployment. These labels appear at the top of the report. Public-facing? drives the bot-factor multiplier and informs sensible daily-cap recommendations downstream.

Federal AI deployments cost more than commercial. Two kinds of overhead apply: multipliers (FedRAMP/GovCloud premium and multi-region redundancy multiply your LLM compute cost), and additive line items (ATO, egress, log retention, vector DB hosting). Set to "none / single / 0" if you're modeling a commercial deployment.

Hosting multipliers

Multipliers stack — e.g. FedRAMP High + active-active = ×2.60 on LLM and GPU costs.

Additive monthly costs

Not modeled: ATO assessment labor (use Agent engineering or Personnel for that), sole-source procurement overhead, GSA Schedule discounts, model availability constraints (some Anthropic/OpenAI models aren't in GovCloud). Add fixed-cost approximations as line items in Infrastructure below.

A "token" is roughly ¾ of a word (so 1,000 tokens ≈ 750 English words). The cost depends on how big each question and answer is. Don't know your numbers? Use the wizard below — it asks plain-English questions and figures out the token math for you.

Different questions have different token cost. A simple lookup uses fewer tokens than a full multi-tool pipeline. Each shape's factor scales the typical query above (1.0 = same size; 0.5 = half). ⓘ

Many real deployments make several LLM calls per user question — e.g., a Planner decides which tools to call, a Retriever ranks documents, a Tool agent formulates a database query, a Summarizer writes the answer. Each agent has its own token budget. If you fill this in, the engine uses agent-sum instead of "Your typical question" above. Leave empty if your deployment is single-call.

Each preset is a different traffic distribution — e.g., "worst case" treats every question as full-pipeline; "mixed" reflects realistic production traffic. Weights must sum to 1.

Break your audience into segments — e.g., "internal staff" vs "public visitors". Each has a different usage pattern. Anonymous public segments get a "bot factor" multiplier to account for crawlers and abuse.

Reserved API capacity (Azure PTU, AWS Bedrock provisioned, OpenAI Enterprise commit) buys a discount in exchange for commitment. Often 30–50% effective savings at >$10K/mo on-demand spend. Choose the plan + size in units (PTU-style) — the engine swaps in the right cost model. Requires API hosting.

Edit the rates / discounts in the Prices tab → API reservations table. Effective monthly cost shows in the report's "Reservation savings" row.

RAG systems pay for embeddings twice: once at ingest (chunking the corpus into vectors, amortized over the re-embed cycle) and once at query time (embedding the user question to do vector search). For a 100M-token corpus + 1M queries/mo this often runs $50–$500/mo. Skip if not using RAG.

Real procurement decisions span phases — "API year 1, hybrid year 2, self-host year 3". Each phase has its own hosting + reservation. Engine computes monthly cost per phase and total 3-year TCO. The chart in the report visualizes cost-over-time so you can see where transitions break even.

Federal RFPs require fully-loaded labor in the cost basis. Each role: FTE allocation × annual base × total-comp multiplier (1.30 = +30% benefits/overhead) ÷ 12. Set FTE to 0 to skip a role. Salaries are editable in the Prices tab → Personnel. This is ongoing post-launch staffing; upfront design effort is in Agent engineering below.

Effort to design, build, and validate the agent before launch — SME interviews, requirements/spec, evaluation criteria, prototype, calibration runs. Roles × FTE × duration. Methodology-agnostic; defaults approximate a CARE-style engagement. Amortized over deployment lifetime so it shows up as a monthly line item alongside Operations and Personnel. Ongoing post-launch staffing lives in Personnel above.

Roles × FTE during design phase

Each row uses the same loaded-salary math as Personnel. Edit role salaries in Prices → Personnel.

Maintenance — re-engineering cadence

Bot rate-limiting protects against runaway cost from anonymous traffic. (The old $/day spending cap was retired — it distorted the cost math by clipping queries at an artificial threshold. To answer "what scale fits my budget?", use the Budget solver section above.)

A catalog of GPU specs the engine can use. The currently-selected GPU is marked with a ●. Hosting strategy is set in Deployment posture at the top.

Inverse of the calculator: enter a target monthly budget and the engine solves for the maximum MAU the deployment can support at the current cost-per-query. Plus what tradeoffs (cache, model swap, batch tier) would unlock more headroom.

Re-runs the full calculation against every available model — same workload, same hosting, same infra, same federal multipliers. Sorted cheapest to most expensive. Δ shows annual difference vs your current pick.

⚠ Cheaper ≠ better. Smaller models (gpt-5-nano, haiku-4.5) cost a fraction of flagship models but lose accuracy on multi-step reasoning, RAG faithfulness, and tool use. Validate any switch against your own evaluation set before committing — savings here are only meaningful if the cheaper model still meets your quality bar.

Each row perturbs one input around its baseline and shows the resulting monthly cost. Sorted by impact (biggest driver on top). The black tick on each bar marks the baseline; red extends to the low case, green to the high case. Useful for budget defense: if a 20% MAU miss blows past your budget ceiling, you need either a larger envelope or stronger optimization levers before signing.

Perturbations: MAU ±20%, Cache hit rate ±10pp, Provider rates ±15%, Bot factor ±20%, Turns/session ±20%. Cache and rate perturbations approximate the deltas analytically — re-run with your own scenario knobs for exact figures.

Projects the monthly headline forward using the Growth/month slider in the simulator. Shows monthly cost climbing month-over-month plus the running 36-month cumulative. Useful for sizing 1-year and 3-year budget envelopes.

Compounds the current simulator growth rate monthly. Caps at 36 months since further projections are unreliable. Headcount, seat-license, and contract reservations don't grow proportionally — adjust manually for those.

Pick two scenarios. The calculator runs each through the same engine and shows the inputs and outputs side by side, with the % difference highlighted. Current = your live scenario in this calculator.

Enter what you're paying today (or what an incumbent vendor quoted you). The calculator surfaces the delta vs the proposed deployment: savings or overrun, plus the payback window if there's a one-time migration cost.

Fixed monthly costs that show up regardless of LLM hosting choice — RDS, S3, CloudWatch, ALB, NAT Gateway, Route 53, etc. Usually the smaller part of total bill but they add up. Each line accepts flat $/mo, $ per query, or $ per GB scaling.