Cost Per Task Estimator
Common agent tasks with estimated token usage. Adjust the model and frequency to see what each one costs per day and per month.
Token counts are estimates based on typical task content. Actual costs vary with real content length.
Email triage
mediumRead inbox, categorize messages, draft replies
3,000 in / 800 out per run
Daily
$0.0840Monthly
$2.52/moPR review
highRead diff, analyze code, post review comments
8,000 in / 2,000 out per run
Daily
$0.5400Monthly
$16.2/moHeartbeat monitoring
lowPeriodic agent check-in and task scan
6,000 in / 200 out per run
Daily
$0.2688Monthly
$8.06/moContent writing
mediumBlog posts, social media, documentation
2,000 in / 4,000 out per run
Daily
$0.0660Monthly
$1.98/moData pipeline
mediumScrape, transform, and report on data
10,000 in / 3,000 out per run
Daily
$0.1500Monthly
$4.50/moCustomer support routing
lowCategorize tickets, assign priority, draft response
1,500 in / 300 out per run
Daily
$0.0480Monthly
$1.44/moSlack/channel summary
mediumRead channel history, summarize key threads, flag action items
12,000 in / 1,500 out per run
Daily
$0.1755Monthly
$5.27/moMeeting notes
mediumTranscribe and summarize meeting, extract action items
15,000 in / 2,000 out per run
Daily
$0.0750Monthly
$2.25/moTotal daily cost
$1.407/day
Monthly estimate
$42.2/mo
vs. running everything on Opus ($277/mo)
saving $235/mo (85%)
Estimates use flat per-task token counts. Real costs vary with actual content length. Use the calculator for more precise numbers, or read the heartbeat deep dive to understand where the tokens go.
How the estimates work
Email triage
3,000 tokens in: inbox summary, email headers, categorization rules. 800 tokens out: labels, priority flags, draft reply snippets. Runs well on Sonnet. Haiku works for simple filtering.
PR review
8,000 tokens in: diff content, context from related files, review guidelines. 2,000 tokens out: inline comments, summary. Complex code needs Opus. Simple style/formatting PRs can use Sonnet.
Heartbeat monitoring
6,000 tokens in: workspace context loaded every check. 200 tokens out: status, any flagged items. This is the high-frequency task. 48 checks/day × context loading = most users' largest cost. Haiku or GPT-4o mini are the right models here.
Content writing
2,000 tokens in: brief, style guidelines, references. 4,000 tokens out: actual content. Note the high output ratio. Output tokens cost 3 to 5x more than input. Content tasks get expensive fast on Opus.
Data pipeline
10,000 tokens in: raw scraped data, transformation rules, schema. 3,000 tokens out: structured output, report. High input volume. Sonnet handles structured transformation well. Only escalate to Opus for complex pattern recognition.
Support routing
1,500 tokens in: ticket content, routing rules, category list. 300 tokens out: category, priority, assigned team. Short, structured, high frequency. This is the textbook Haiku or GPT-4o mini case.
Need a model recommendation?
Tell the routing engine what workloads your agent runs. It will recommend models for each task and output a config snippet you can paste into openclaw.json.