Trim Playground
Agent Economics
Inputs
Compare models →Model
$0.002500/1K in · $0.0100/1K out · $0.001250/1K cached in
Workload
Tokens per LLM call
Pricing modifiers
Bills some input tokens at a cached input rate (if available).
Cached share
%
Applies a multiplicative factor to token cost (set it to your provider’s batch discount).
Batch factor
Use 1.00 if you don’t have batch pricing.
Uncertainty (optional)
This is an assumption multiplier for retries, multi-call orchestration, context growth, and tool costs. It’s not derived from a single reliable public dataset.
Results
LLM calls / month
100.0K
Selected monthly cost
$600.00
(with modifiers: caching off, batch off)
Scenario comparison
Base (no modifiers)
120.00M tokens / month
$600.00
$0.006000 / call
Caching only
120.00M tokens / month
$600.00
$0.006000 / call
Batch only
120.00M tokens / month
$600.00
$0.006000 / call
Caching + Batch
120.00M tokens / month
$600.00
$0.006000 / call
Per-call breakdown
Input: 800 tokens → $0.002000
Output: 400 tokens → $0.004000
Total: $0.006000
What this excludes
- Tool/API costs (search, DB, browser, compute)
- Retries on failures and self-correction loops
- Context growth across multi-step tasks
- Computer-use / GUI action overhead
Notes
Use Tasks/month + LLM calls/task for agentic workflows. If you only know “requests”, keep it on Requests/month.
Prompt caching is modeled as a share of input tokens billed at a cached input rate (when available in the model config).
Implementation lives in src/lib/calculator/agenticCost.ts.