OpenWeightsTerminal

Earn

Estimates, not guarantees. Inference earnings are in US dollars. Network rewards are points, shown separately. Change any assumption below.

Inference earnings (estimate, USD)

Callers pay for the tokens your box generates, at your price per million. You keep 80%. These numbers only hold if paid demand reaches your node.

Assumptions

Default: market reference price for this family.
15%
Default 15%, a conservative guess. It depends on caller demand, not on your box.
Host share80%

Per box: ServeTheHome measured 60–90 W serving LLMs and 40–45 W idle (240 W power supply). We use the top of each range; several concurrent streams may draw more. Two-box setups count two.

Desk US street price per box, Sep 2026 (Amazon US floor). NVIDIA Founders Edition list is $4,699. Two-box setups count two.
Callers pay$14.73per month
Your share (gross)$11.78per month
Power cost−$5.6738 kWh per month
Net$6.12per month
Net per year$73.40per year
Payback68 yrestimate
Annual return on hardware (inference only)1.5%Estimate. Net × 12 ÷ device price. Points are not included.
tokens/mo = tok/s × 3,600 × 730 h × utilizationcallers pay = tokens/mo ÷ 1M × pricegross = callers pay × 80%power = (util × load W + (1 − util) × idle W) ÷ 1,000 × 730 h × $/kWhnet = gross − power; payback = device price ÷ netannual return = net × 12 ÷ device price

31.3M tokens a month at these settings. Counts generated tokens only. Callers are also billed for prompt tokens, so real billing can be higher; it never assumes more demand than the utilization you set.

Network rewards (points)

Two kinds of points. Usage points: 1 point per $0.001 of paid caller spend on an eligible node (1.5× on priority models), capped at 1,000 a day. Bootstrap points: 100,000 a week shared pool-wide, scaled down as paid revenue grows, split by time online. Usage points only accrue on models in the reviewed model program (see the host catalog on Pricing); other models earn dollars and bootstrap points only.

Points have no cash value today. Any future token is not promised, and points are not a return on your hardware.

Assumptions

Including yours, at similar uptime. Default 25.
The bootstrap pool is 90% of 100,000 points below $200 a week and falls to 10% at $2,000.
100%
Bootstrap points are split by status checks passed while online.

This model is not in the reviewed model program, so it earns no usage points. Bootstrap points still apply.

Usage points0per month
Bootstrap points3,600per week
Bootstrap points15,643per month
Total points15,643per month
usage = callers pay × 1,000 × multiplier, max 1,000 a daybootstrap pool = 100,000 × 90% a weekyour bootstrap = pool × your uptime ÷ (your uptime + other nodes)

What it takes to earn

The desk measures every node. A node earns points only while it passes every bar below, and earns dollars only when callers send it jobs.

  • Throughput of at least 44.9 tok/s on this model: 60% of the desk benchmark (74.8 tok/s), never below 15.
  • At least 95% of jobs finish successfully.
  • Online at least 80% of the last 24 hours.
  • At least 5 finished jobs before speed and success are judged.
  • A passed endpoint check within the last 24 hours, and a desk test job finished each week for bootstrap points.

Earnings depend on caller demand. Jobs go to nodes serving the requested model at a price callers accept, so a fast box with no callers earns nothing.

Host and earn →