BREAKING
Run Claude Fable 5 at 1/10th cost
0
$/1M
input tokens
0
$/1M
output tokens
0
M
context
0
x
tokens vs Opus 4.8
0
%
input cache discount
0
k
max output
Multi-tier routing strategy
1
Classify with Haiku
↓
2
Glue work on Sonnet
↓
3
Reserve Fable 5
↓
4
Add prompt caching
0
%
higher detection rate
0
%
sessions fall back to Opus
Routing becomes standard practice
AI NEWS BLITZ
Developers are sharing a playbook to run Claude Fable 5 at up to one-tenth the cost.