BV Compute Tokens
Four instruments covering the compute stack from input to outcome. All four are testnet.
Four instruments covering the compute stack from input to outcome. All four are testnet. Nothing on this page is available on mainnet.
Input, runtime, outcome, and the IP layer beneath them.
BVGPU
The input leg.
Testnet. One verified hour of qualified GPU compute (H100 / A100 / equivalent), provider-attested.
GPU-hours are the input metric: what you buy when you are buying capacity rather than a result. See BVGPU for the full instrument documentation.
BVINF
The runtime leg.
Testnet. One verified inference call against a registered model, with per-call attestation.
If BVGPU is capacity, BVINF is consumption — priced per call rather than per hour. See BVINF.
BVMDL
The IP leg.
Testnet. Licensed access to a specific model weight set or fine-tune, with usage metering and a revenue share to the original creator on downstream inference.
BVMDL is a different kind of instrument from the other three. Where BVGPU, BVINF and BVOUT are capacity or delivery units, BVMDL conveys a contractual IP licence together with a downstream royalty on inference. Its rights statement is its own.
See BVMDL.
BVOUT
The outcome leg.
Testnet. A verified, deterministic compute result — a finished training run, a delivered batch prediction set, a completed simulation.
Buy the outcome rather than the hours. See BVOUT.
A note on BVNRG
BVNRG belongs to the energy suite, not this family — long-dated wholesale power
forwards, documented separately at BVNRG.
More from BV Compute
MXTK API Gateway
TestnetApollo Federation GraphQL gateway + REST surface. Single query surface across mortgages, tranches, energy tokens, tax credits, compute markets.
BVGPU
TestnetGPU-hour token. Each BVGPU represents one verified hour of qualified GPU compute (H100 / A100 / equivalent), contractually eligible for physical delivery or cash settlement subject to agreement terms.
BVINF
TestnetInference token. Each BVINF represents a paid inference call against a registered model, settled in MXD with per-call attestation.