Tool
AI cost per task calculator
Compare two models on what a task really costs: the tokens, the time people spend reviewing, and the time they spend fixing mistakes. The cheapest model per token isn’t always the cheapest per task.
Claude Fable 5.1 is cheaper: $4.18 less per task (72%), about $8,352 a month at 2,000 tasks.
Model A · Claude Fable 5.1$1.64 per task
- Model
- $0.750
- Review
- $0.000
- Fixing mistakes
- $0.891
- A month
- $3,282
Model B · Claude Haiku 4.5$5.82 per task
- Model
- $0.075
- Review
- $0.000
- Fixing mistakes
- $5.74
- A month
- $11,634
ModelReviewFixing mistakes
How it works
A cheaper model per token can cost more per task once people review its work and fix its mistakes. The calculator adds all three. The math and examples come from our guide to AI model cost per task.
| Tier | Best for | Examples | Price per million tokens, in / out |
|---|---|---|---|
| Frontier | Hard reasoning, ambiguous judgment, long multi-step work | Claude Fable 5.1, Claude Opus 5 | $10 / $50 and $5 / $25 |
| Mid-size | Most everyday drafting and analysis | Claude Sonnet 5 | $2 / $10 |
| Small | Classifying, extracting and routing at volume | Claude Haiku 4.5 | $1 / $5 |
| Open-weight | High volume, data that must stay in your network, custom tuning | Families such as Qwen, GLM, Kimi and Llama | Low per token, plus the cost of hosting and running it |
Questions people ask
How is cost per task calculated?
Model cost per attempt times attempts, plus review time times people’s hourly cost, plus the share of mistakes times the time to fix one times hourly cost.
Where do the prices come from?
The presets use Anthropic’s published list prices per million tokens, accessed October 2, 2026. Choose “Custom price” to enter any model’s prices, including open-weight models you host.
Where do I get accuracy numbers?
From your own evals: run each model on real cases from your work and count how often it gets the task right the first time. Public benchmarks rarely match your work.
Why include people’s time?
Because mistakes cost more than tokens. A small model that is wrong more often can cost more per correct result than a frontier model, once someone has to fix its work.