A token backed by hardware we own.
Not a yield instrument — a claim on real compute we own. A subscription is just a pack of $CMP: spend it as you go, or stake it and draw inference forever.
Spend
Pay as you consume. Inference draws down your $CMP per the spend table. Best for light or bursty use — the price is fixed and there’s no lock-up.
- GLM-5.2 output ≤32K250 /M
- GLM-5.2 input50 /M
- long-context output375 /M
- fast-small model90 /M
Stake
Lock $CMP instead of spending it. Each day it grants you a Diem — a day’s usage allowance (Latin for “day”) sized to your share of total stake — forever, while the principal stays yours.
- daily allowance = stake ÷ total × pool
- pool grows as we add rigs — no token printing
- Diem is use-or-lose (36h) — a usage right, not yield
One $200 pack, two ways
~40 messages/day. Burns ≈330 $CMP/month.
Stakes the 22,500 instead of spending it (at target TVL).
The crossover is ~2,500 $CMP/month of usage: below it you rationally spend, above it you rationally stake. As more heavy users stake, each share shrinks and the crossover rises — so the economy self-balances instead of everyone staking and starving cashflow.
Why it doesn’t collapse
Capacity governor
The staker pool is capped at ≤35% of sustainable throughput. Total Diem claims can never exceed real compute — overselling is structurally impossible.
No promised yield
Diem floats with measured spare capacity and expires in 36h. There’s no fixed APY to break and nothing that reads as a security.
Backed by growth
Add rigs → the pool grows → every staker’s allowance rises, with zero token emission. Value comes from real compute, not inflation.
Stress-tested against mass-unstake, demand spikes, price crashes and low utilization — stable within the parameter envelope above.
Buy compute, not a subscription.
Free tier is open now. Early users get a $CMP allocation at launch.
Open the terminal