How much could you save?
Pick the models you run, or drop in a usage export, and see the same work priced on open-weight models with Pearl.
$81.53
35% lower cost$231.85 on your providers → $150.32 on Pearl
per month, 100M tokens on each of 4 models
Models I use
Tokens per model, per month
Cache hit rate
1% in · 1% out
But why guess?
Drop your billing details below and the comparison switches to your real usage.
Providers
Anthropic
OpenAI
Cursor
Files are parsed in your browser and never uploaded.
xAI
Grok 4.6
$14.34
25% lower · per month
| Reasoning effort | Comparable model | Intelligence | Savings |
|---|---|---|---|
| high• | Kimi K3@max | 60.9→59.7 | $14.34 (25%) |
Each row is the same month run at that reasoning effort. The dotted row is the one behind the totals.
1 / 4
Estimates, not quotes. Your month is modeled as 100M tokens on each selected model, split 98% cache reads / 1% fresh input / 1% output, with no cache writes (a hit rate says nothing about how the cache filled). Each model is priced at its provider's list rates (snapshot 2026-08-16) and matched to the open-weight configuration whose Artificial Analysis Intelligence Index sits closest (snapshot 2026-08-16); ~ marks an estimated index. Both sides price the SAME token volume at list rates, Pearl's at 98% cache reads too (2026-08-10 snapshot, 10% below Together AI serverless), so a model that needs more tokens to finish the same work would cost more than shown. Totals use each model's headline configuration; the other rows are the same month at a different reasoning effort. Upload an export to replace all of this with your real usage.