What will your agent cost?
Plan the full cost of a task: model calls, retries, tools and the time you spend checking the result.
Estimates update in your browser. All amounts are in USD.
Examples are made-up starting points, not measured usage. They replace the task inputs and keep your model.
View estimate · $244.201. Your task
$1.00 input / $5.00 output per 1M tokens. Pricing source
Original tasks, before any retries.
Include every step and loop in one attempt.
Average across calls, including instructions, history and tool results. Up to 200,000.
Include billed thinking or reasoning tokens. Up to 32,000.
2. Retries and other costs
10 means 10 extra full attempts for every 100 original tasks.
Your expected final success rate after retries. A planning assumption, not a model benchmark.
Total search, browsing and other tool fees for one full attempt. Retries repeat these fees.
Only the share of hosting and subscriptions used by this task. Avoid counting shared costs twice.
3. Human work
Checking and fixing results takes time. Set the share to zero if you want to estimate bills only.
Share of original tasks reviewed, including failures when needed.
Total review and correction time for each reviewed task, including any repeat checks.
Use your own hourly cost. This is the value of time, not automatically an extra bill.
Compare monthly estimates
Each total uses the same task and human time. Quality and review needs may change, so check with a small trial.
Pricing sources and what the estimate excludes
The listed models use standard, paid text API rates in USD, checked 2 October 2026. Caching, batch discounts, free credits, regional surcharges and taxes are excluded. Setup and maintenance time are excluded. Add recurring maintenance to your review time or fixed costs when needed. Chat subscriptions and API usage have separate terms.
This calculator supports up to 200,000 input and 32,000 output tokens per call. Use billed output tokens, including thinking. Average usage can hide large calls or growing conversation history. Check the actual limits and charges for your provider. Custom prices let you use a different model or your own contract.
- OpenAI · GPT-6 Luna pricinggpt-6-luna
- Anthropic · Claude Haiku 4.5 pricingclaude-haiku-4-5-20251001
- Anthropic · Claude Sonnet 5.5 pricingclaude-sonnet-5-5
- Google · Gemini 3.1 Flash-Lite pricinggemini-3.1-flash-lite
Does this tell me which agent is best?
It estimates cost from your inputs. A low token price does not show whether a model can complete your task. Test a small sample, check its results, and update the calls, tokens, review time and success rate before you plan a larger rollout.
How we calculate the cost
One task can take several model calls. An attempt includes all those calls. Extra attempts repeat the same average model and tool costs. Human review uses the share of original tasks you select and the total review time per task.
- Attempts per month
- Tasks × (1 + extra attempts per 100 tasks ÷ 100)
- Model cost per attempt
- Calls × (input tokens × input price + output tokens × output price) ÷ 1,000,000
- Monthly bills
- Attempts × (model cost per attempt + tool fees per attempt) + fixed monthly costs
- Value of human review
- Tasks × review share × review minutes ÷ 60 × hourly rate
- Budget per successful task
- (Monthly bills + review time value) ÷ (tasks × final success share)
Find an agent for your task
Turn a cost estimate into a small, useful trial.