AI Operating Cost Estimator

Simulate AI operating cost at scale.

Configure workflows, volume, and model
to simulate AI operating cost.

Simplified inference-based simulation using configurable assumptions and live model pricing.

1
Your use case
2
Choose a model
3
Your estimate
Organisation size preset

Sets daily volume across all tasks based on organisation size and estimated adoption rate.

Size AI users Uses/day Adoption rate Runs/day per task
1–20 people124~80%48
20–100 people455~75%225
100–500 people2006~70%1 200
500+ people4507~65%3 150
Assumes active AI deployment across the organisation. Adjust individual task volumes below if your adoption differs.
Select tasks & set daily volume

Select which workflows to include and adjust volume per task. One run = one AI request.

Token estimates are directional — based on typical enterprise usage patterns.
Cost is driven by:
· Prompt size and document/context volume · Number of API calls per workflow trigger · Workflow complexity and orchestration depth · Automated workflows generate significantly higher token usage than single AI requests
Everyday tasks — single API call per run
Task Input Output
Summarise document~8k~600
First draft~600~1.2k
Email drafting~400~350
Answer question~4k~500
Invoice processing~3k~400
Translate~3k~2.8k
Screen / filter~5k~300
Feedback analysis~2k~600
Automated workflows — multiple API calls per trigger
Workflow API calls Total tokens
Simple workflow3–4 calls~18k
Complex workflow8–10 calls~80k
Developer tasks — single API call per run
Task Input Output
Code generation~5k~1.2k
SQL query~800~400
RAG Q&A~6k~700
API documentation~4k~2k
All input estimates include system prompt overhead (typically 200–800 tokens/call). Actual costs vary with prompt design, document length and retry frequency.
This simulator models active workflow-driven AI usage and inference consumption. Seat-based AI subscriptions (e.g. Copilot, ChatGPT Team, Gemini Workspace) may include bundled usage and are therefore not fully reflected in inference-based estimates.
Working days / month
Working days 22
122 (typical)31
Model
Model selection affects inference cost only. Output quality, latency and data residency vary by provider.
Lowest cost model for this volume
Automatically selects the cheapest available model for the selected workload. Always shows cost vs selected model. No capability or quality evaluation.
—
—
Optimisations
Batch API −50% input cost
Async batch processing — non-real-time tasks
Prompt caching off
How much of your input is static (system prompt, document base)? The higher the share, the more caching saves.
0% static
Drag right to model how much of your input context is reused across requests.
Sensitivity parameters

Adjust real-world factors that affect token consumption beyond base estimates. Agent and complex workflow costs are most sensitive to these parameters.

Retry rate 0%
How often workflows fail and re-execute. Each retry repeats full token usage.
Orchestration depth 1×
Token multiplier for agent steps. A 3-step agent re-reads prior context at each step, amplifying token use.
Context growth 1×
How much context accumulates in complex agentic workflows. Higher values model longer chains and tool outputs.
Compare two models

Side-by-side cost comparison of two models. Price-only scenario modelling. Opens OpenRouter for deeper benchmarking, including context window, latency, and capability data.

Model A
vs
Model B
Productivity impact estimate
What level of workflow efficiency improvement do you expect from AI adoption?
Active AI users 0 people
Avg. fully-loaded cost per person kr / yr
Estimated efficiency improvement 10%
5% — limited assistance20% — integrated into workflows40% — highly automated
Cost over time
Projected monthly cost as AI usage grows
Expected monthly growth
⏳ Loading prices and rates…
Monthly cost estimate
kr 0
Model Claude Sonnet 4.6
Active AI users —
Workflows —
Runs / day —
Working days 22
Batch off
Cache off