📊 Not sure of a value? Fields with a ▾ Typical pill offer broad industry ballparks (sourced typical ranges; Token shapes mirror the site use-case profiles (chatbot/RAG/agent/coding) - the same shapes the model finder uses.) so you can move forward now — your result gets more accurate as you replace them with your own measured numbers. Values marked * are rough estimates.

🎮 Interactive Guide & Calculator Playground →

Size the workload

Enter tasks/day and tokens/task — these set the raw volume.

Pick model + savings

Choose a model tier and set cache hit rate and batch-eligible share — the big discounts.

Read the monthly cost

You get a realistic monthly figure with a safety buffer for runaway loops.

👇 Now try the calculator below with your own AI workloads

🔗

Workflow type

Sets tasks/day, tokens/task, model tier and (in Advanced) cache, batch, and runaway buffer.

Agentic tasks per day (across team)

Tokens per task (avg)

Model tier

Catalog freshness:

✓ Curated · verified today ago

Estimated monthly cost

$66/mo

~$791/year · 1,100 tasks/mo

DeepSeek

$31/mo

DeepSeek V3.2 (reasoner)

OpenAI

$60/mo

GPT-5.4 Nano

Google

$72/mo

Gemini 3.1 Flash-Lite

Anthropic

$249/mo

Claude Haiku 4.5

Vendor choice matters here - the spread between cheapest and priciest is 706%. Routing the right tasks to the right model can cut your bill significantly.

Workflow scale

Tasks per day (across team) Hint: Jobs your team runs through this per day, everyone combined.

Tokens per task (avg) Hint: Words in + out for one full job. Short ~2,000, heavy ones tens of thousands.

Reference: small refactor ~25K · Claude Code session ~100K · multi-file research ~300K · long autonomous run ~800K

Working days per month Hint: 22 for business-days-only; 30 if it runs every day.

Token split - input share (0.1–0.95) Hint: Share that is input vs the reply (0.7 = mostly input). Leave as-is if unsure.

0.7 = 70% input / 30% output (typical code-editing). Lower for output-heavy workflows.

Model & optimization

Model tier Hint: Cheaper tiers handle most work; premium only for the hardest reasoning.

Cache hit rate (0–1) Hint: How often the same setup text is reused (0-1). Higher = cheaper. 0 if unsure.

Fraction of input tokens served from cache (system prompts, repeated context). Vendors without published cache rates get 0 regardless.

Batch-eligible share (0–1) Hint: Share of jobs that can wait (0-1). Those run about half price.

Fraction of workload that can run async (50% discount). Most realtime agents = 0.

Runaway-loop safety buffer (1.0–3.0) Hint: Safety padding for agents that loop too long. 1.5 = budget 50% extra. Leave default.

Multiplier for unexpected agent loops. 1.2 = 20% buffer (recommended).

Per-vendor breakdown

Vendor / Model	Cached input	Realtime input	Output	Monthly
DeepSeek DeepSeek V3.2 (reasoner)	$1	$13	$17	$31
OpenAI GPT-5.4 Nano	$1	$9	$50	$60
Google Gemini 3.1 Flash-Lite	$1	$12	$59	$72
Anthropic Claude Haiku 4.5	$5	$46	$198	$249

Want to model this as an architect? Try the Agent Loop Cost calculator for per-task breakdown, or Agentic AI Stack for planner+executor stacks. For full TCO, run the TCO Wizard. See methodology →

Open full calculator on aicost.ai →

Agentic Workflow Cost Calculator

Estimate your monthly agentic-AI burn

Results

Workflow scale

Model & optimization

Per-vendor breakdown