AI Agent & LLM Cost Calculator

Answer a few plain questions and get a clear, costed model recommendation — no jargon, no token counts. Power users can switch to the full calculator any time.

Step 1 of 5

👋 Start here — three steps to a model choice

No token knowledge needed. Every input has a “Not sure?” helper.

Pick what you're buildingTap a preset below — it fills a complete, believable scenario.

Adjust the 3 basicsRequests/month and reply length are what move the answer. Leave “Fine-tune” alone unless you know it.

Read your Bottom lineThe card below names your cheapest pick, a premium alternative, and your biggest saving.

How to read this dashboard — tap a step to jump there

Your workload — one click sets a full scenario, then tweak

Basics — these drive the answer

Uses / month (requests)

Typical: a small app 5K–50K · a busy product 100K–1M+. This scales every cost linearly.

Text you send (input tokens)

What it is: the prompt you send (system + context + question). Typical: 500 (simple) to 15,000 (document/RAG).

Reply length (output tokens)

What it is: how long the model's reply is. Output is usually 3–5× pricier than input, so this matters most. Typical: 150 (a sentence) to 4,000 (an essay).

Fine-tune (optional) — caching, reasoning, batch, growth

Thinking effort (reasoning tokens)

What it is: hidden “thinking” the model does before answering — billed at the output rate. Typical: 0 (simple), 500–2,000 (planning/agents).

Reused text savings (cache hit) — 0%

What it is: the share of your input the model has seen before and can reuse cheaply. Typical: 40–70% for agents with a fixed system prompt; ~0% for one-off calls.

Cache read = input × 0.10

What it is: how cheap a cached read is vs a fresh one. 0.10 = cache costs 10% of the input price (Anthropic & OpenAI default). Range: 0.05–0.30. Leave at 0.10 if unsure.

Batch discount

50%

What it is: a discount (%) for jobs you don't need answered instantly — evals, back-fills, bulk summaries. Typical: 50%. Toggle on + set the %; applies only to providers that offer batch.

Monthly growth — 0%

What it is: how fast your monthly usage rises, used by the 12-month projection only. 0% = flat. Typical: 5–10% steady, 15–25% for a fast-scaling product.

Monthly cost by model ✓ = cheapest · ● = your pick

What you're paying for

12-month projection — compounded by the growth slider

How we calculate this: month 1 = your current monthly cost; each later month multiplies usage by (1 + growth%). The total is the sum of all 12 months.

Cost as you scale — monthly cost vs usage, your volume marked ?

What changes the answer most — your biggest cost levers ?

How we calculate this: each lever is reduced by 20% (cache hit is raised by 20 points) and we measure the change in monthly cost for your chosen model, then rank by dollar impact.

Model comparison — click a row to set your pick · tick up to 4 to compare

Vision only 128K+ context Max $/mo

Cheapest lowest cost · Best value cheapest fully-capable · Premium frontier reasoning · Largest context | ✓ verified · est estimate (editable).

Multi-agent simulator — orchestrator + workers, each on its own model ?

Each row is one agent. Reqs/mo = how often it runs · Text in = prompt size it gets · Reply out = answer length it produces. The system total sums every agent's monthly cost.

System total: $0 / mo

Break-even analysis ?

Model A

A fixed $/mo

Leave 0 for pure usage pricing. Add a number to compare managed vs self-hosted.

Model B

B fixed $/mo

Same idea for model B. Different fixed costs are what create a real break-even point.

How we calculate this: the monthly volume where A's usage cost + fixed cost equals B's. Below it the lower-fixed-cost model is cheaper; above it the lower-per-use model wins.

ROI estimator ?

Value per task ($)

What it is: the money one successful task is worth to you (saved support time, a drafted email, a closed lead). Typical: $0.20–$5. Only affects this ROI table.

Model	Cost/mo	Value/mo	Net/mo	ROI

How we calculate this: Value/mo = value per task × your monthly uses. Net = value − cost. ROI = net ÷ cost.

Rate-limit check ?

Speed limit (requests/min, RPM)

What it is: how many requests/minute your account is allowed. Find it in your provider dashboard. Typical: 500 (free) to 10,000+ (scale).

Throughput limit (tokens/min, TPM)

What it is: how much text/minute (in + out) your account can process. Typical: 200K (free) to 10M+ (scale). Leave blank to skip the check.

📖 Glossary — what every input means & what it changes

Input	What it means	Why it affects cost	Typical	Changes

Save & compare scenarios — e.g. “Current” vs “Optimised”

Snapshots store your workload, prices, batch/cache settings and primary model in this browser (localStorage).

Capability matrix — published values only; unknowns left blank

Context windows and modality flags reflect providers' published docs as of October 2026. Per-account RPM/TPM are tier-specific and omitted — set yours in the Rate-limit check. Blank = not publicly confirmed (not “unsupported”).