01↗
Legible
Every response carries the watt-hours it burned, the dollars that came to, and the share of your prompt served from cache. Cost stops being something you reverse-engineer from an invoice.
Reported per requestEnergy-priced inference API
One rate for every model.
Billed in kilowatt-hours.
Traversaal Energy is an inference API that bills the energy your requests consume. Every response comes back with what it used, in watt-hours and in dollars.
$10/kWhFlat, every model
88%Under token rates
0.22sMedian time to first token
This request
glm-5.2 · flat rateEnergy consumed94.9mWh · $0.0009
Requests metered32 / 32 billed
↓ 80%
vs token priceSame request
The idea
Same rate for a 31B
and a 400B.
Per-token prices bake each model’s markup into the unit itself, so a bigger model costs more per token by decree. A kilowatt-hour costs what a kilowatt-hour costs. Price on that and an efficient model becomes genuinely cheaper to call, not just cheaper on the rate card.
See it on a live request ↓Industry first
Inference priced by
the kilowatt-hour.
02↗
Flat
One rate across the whole catalogue. Moving from a 31B to a 400B changes how much energy you consume. It never changes what you are charged per unit of it.
No per-model markup03↗
Rewarding
Trim a prompt, raise your cache rate, switch to a sparser model, defer batch work to the flex tier. Under energy pricing every one of those lands on your bill instead of our margin.
Optimisation pays youTry it
Send a prompt. Read the meter.
Switch models and watch both numbers move: what the request cost in energy, and what the identical call would have cost per token.
Start a conversation. Every response comes back with the watt-hours it consumed and what that cost.
Preview only. Completions are canned and served from the browser, not a live endpoint. Token prices are the posted rates for these models; energy figures are modelled from measured consumption under batched serving, at a 31% cache hit rate. Real requests meter their own energy.
Estimate
What your month looks like, metered.
Put your volume and your current token rate side by side. The energy line is metered kilowatt-hours times the flat rate, and nothing else.
- 01 Set your monthly token volume
- 02 Enter the token rate you pay now
- 03 Compare against energy pricing
Estimate your bill
$10.00 / KWHEnergy pricing, same usage
$108 / mo$70.20 on flex80% cheaper$540 on tokens
≈ 10.8 kWh
metered per monthat 0.09 Wh per 1K tokens
An estimate, not a quote. Energy per token moves with model architecture, request size, cache hit rate, and cluster load. Sparse models land well under this line, long-context reasoning runs above it. Your real bill meters each request.
Why Traversaal Energy
Three things every layer is built around.
Metering you can read, serving that earns its watts, and a rate card that rewards efficiency instead of taxing it.
01Reporting
Energy reporting, on every plan.
Watt-hours and cost on each completion, per-key trends in the dashboard, and side-by-side efficiency across models. Included even if you stay on per-token billing.
02Performance
Serving built for throughput.
Continuous batching, tensor parallelism across GPUs, and prefix caching that skips recomputation on repeated context. 0.22s median time to first token on the fast tier.
03Efficiency
More output per kilowatt-hour.
Sparse mixture-of-experts models activate a fraction of their weights per token, so they meter far below a dense model of the same size, and under a flat rate you keep that difference.
“Every token you buy is a quantity of energy someone else measured, marked up, and rounded off. We would rather just show you the measurement.”
Nothing else changes
OpenAI-compatibleOne endpoint, many modelsStreaming, tools, JSONReporting on every plan
Change one line
Point your base URL at us.
Tell us the models and volume you run today and we will come back with what it meters at.