Live product tour
Connect a read-only provider key and Lisran backfills your last 30 days. This is the real Overview, running on a sample workspace.
Your AI cost across every connected provider.
Last 30 days
$13,531.00
29 Sep – 28 Oct 2026
Daily average
$451.03
over 30 days
Week over week
+9%
$3,515.40 vs $3,230.02 — final week of period
Projected (next 30d)
$15,065.96
▲ 9% vs prior week
11 models in total
Route claude-3-opus → claude-3-5-sonnet where quality allows
Claude 3.5 Sonnet beats Opus on most tasks for far less. You're spending $7,746.01/mo on claude-3-opus.
Save up to $6,196.81/mo if fully routed
Route gpt-4o → gpt-4o-mini where quality allows
gpt-4o-mini handles classification, extraction, and short tasks well. You're spending $1,360.84/mo on gpt-4o.
Save up to $1,279.19/mo if fully routed
Route gpt-4-turbo → gpt-4o where quality allows
gpt-4o matches gpt-4-turbo quality at a fraction of the price. You're spending $1,654.11/mo on gpt-4-turbo.
Save up to $1,163.05/mo if fully routed
Ceilings assume every call moves. Treat them as the upper bound, not a forecast.
Every other breakdown here says where the money went. This one says what it went on — and a task is something you can actually change.
Route to claude-3-5-sonnet — up to $2,610.28/mo
Route to claude-3-5-sonnet — up to $1,108.80/mo
Route to claude-3-5-sonnet — up to $1,565.52/mo
Route to gpt-4o — up to $308.89/mo
Route to gpt-4o-mini — up to $116.68/mo
22% of spend ($2,922.69) carries no task tag — it came straight from a provider, or through the gateway untagged. Those figures are real but absent from this table.
Storage inside a token bill is the cost of keeping a prompt warm instead of sending it again. A read costs a tenth of the input rate; a write costs more than not caching at all, so a cache nobody reads back loses money.
Data held
436 GB
+33% · $1,124.38 this period
Served from cache
17%
6.1 reads per write
Saved by caching
$2,291.41
vs. sending the same tokens uncached
Still uncached
$2,473.70
ceiling if all of it could be cached
Some of this spend is on a provider whose cache rates we have not established, and is priced at the full input rate. These figures are upper bounds.
Read-only usage scope · encrypted at rest · about a minute to your first numbers. Every figure above is computed from a sample workspace, not a real customer.