Coming soon

Routing economics

Compare frontier API costs vs self-hosted open models across traffic tiers

Simple queries60%
$12K
frontier API · Claude 3.5 Haiku
200 in / 100 out tokens per query
↻ configure
60%
Standard queries30%
$92K
frontier API · Claude 3.5 Sonnet
800 in / 400 out tokens per query
↻ configure
30%
Complex reasoning10%
$55K
frontier API · GPT-4o
2000 in / 1000 out tokens per query
↻ configure
10%
Annual savings
$72K
45% less
Self-hosted saves vs frontier API
Frontier API cost
$159K
100K queries/day × 365
Self-hosted cost
$87K
11 GPUs total · hardware amortization
Cost vs. volume
breakeven ~55K/day$0$95K$191K$286K$382K$477K0/day60K/day120K/day180K/day240K/day300K/day
Frontier API Self-hostedBreakeven at ~55K queries/day
ModelTypeTierAnnual costCost / query
Claude 3.5 HaikuAPISimple queries$12.3K$0.000560
Llama 3.1 8BSelf-hostedSimple queries$4.6K$0.000210
Claude 3.5 SonnetAPIStandard queries$92.0K$0.0084
Llama 3.1 70BSelf-hostedStandard queries$16.4K$0.0015
GPT-4oAPIComplex reasoning$54.8K$0.015
Llama 3.1 405BSelf-hostedComplex reasoning$65.8K$0.018
Total
API $159.0K
Self-hosted $86.8K