Break-even index / RTX 5090 / Claude Sonnet 4.6 / District of Columbia
Will a RTX 5090 build beat Claude Sonnet 4.6 in District of Columbia?
Break-even lands around month 29. Real, but slow — a hybrid setup (local for the routine work, cloud for the peaks) usually beats going all-in.
How to read it: OWN — the rig pays for itself inside 18 months. HYBRID — break-even lands between 18 and 36 months; real, but slow. RENT — it never pays off before the hardware is due for replacement. Every formula and threshold is public on the methodology page.
The numbers
| Build cost (RTX 5090 build) | $3,500 ($3,000–$4,000) |
|---|---|
| Hardware amortized (3 yr) | $97.22/mo |
| Electricity in District of Columbia ($0.182/kWh) | $12.29/mo (67.5 kWh) |
| Local total cost of ownership | $110/mo |
| Claude Sonnet 4.6 at the same usage | $135/mo |
| Break-even | about 29 months |
Assumes the "coding assistant" preset: ~15M tokens/month, 3 h/day active, 5 h/day idle. Claude Sonnet 4.6 priced at $3/M in, $15/M out (verified 2026-07-05). Rig ceiling: 70B+ with real context windows. Jargon: VRAM is the GPU memory that decides which models fit; Q4 is 4-bit quantization — a smaller, faster copy of a model with a slight quality cost.
See your verdict — then tweak every number →