Did you know that if you run GLM 5.3 Full on 4x DGX Sparks locally, you'll break even in ~5-6 months?
Assuming:
> you run 10B tokens/mo:
> having 99% input (95% cached), 1% output
Per API Rate:
> 9.4B cached x $0.26/M = $2,445
> 0.5B fresh x $1.40/M = $693
> 0.1B out x $4.40/M = $440
> Total: $3,578/mo
Therefore >> 4x DGX Spark: $18,796 (a few months ago) + ~$60/mo power