Free LLM cost calculator

Calculate cost per task, not just cost per million tokens

Model price is only one input. Compare two workloads using calls per task, actual token shape, cache usage and retry overhead to see what one completed unit of work really costs.

Why this calculator stays current
  • You enter your actual provider or negotiated rates
  • Cached input is priced separately from uncached input
  • Retries and extra agent turns are part of task cost
  • No vendor price table to silently go stale

Workload assumptions

Compare two real task shapes, not two sticker prices

Use the rates from your provider contract and the token/call shape from your own workload. The example values are illustrative, not a live provider price table.

%
Tokens served from a provider cache, entered separately from uncached input.
Use 0 when your provider has no separate cache-write charge.
$/1M
$/1M
$/1M
$/1M
%
Tokens served from a provider cache, entered separately from uncached input.
Use 0 when your provider has no separate cache-write charge.
$/1M
$/1M
$/1M
$/1M

Workhorse

$0.0290
estimated cost per completed task
Effective calls / task
4.20
Cost / model call
$0.006900
Monthly cost
$289.80

Frontier

$0.3548
estimated cost per completed task
Effective calls / task
5.50
Cost / model call
$0.064500
Monthly cost
$3,548

Comparison

Workhorse is cheaper by 91.8%

At 10,000 tasks/month, that is about $3,258 per month.

This is an estimate from the assumptions above. Real cost per task changes with retries, context growth, cache hit rate, routing and task success—exactly why production measurement matters.

Stop estimating after the benchmark

UsageBox can meter the numeric token quantity or cost value your application already computes. Keep richer task/model/eval metadata in your application analytics, or separate usage streams with explicit meters/product items.

Price per token is not task cost

A cheaper model can need more calls, more context or more retries. Cost per task includes that behavior instead of assuming identical token consumption.

Cache hit rate changes the answer

Repeated context can be much cheaper than fresh input. Treat cached input separately instead of applying one blended input price to every token.

Production measurement still wins

The calculator is for planning and comparison. Once traffic is live, meter the quantity or computed cost your application actually produced rather than relying on a static estimate.