The inference engine
for autonomous systems.
Point your agents at one endpoint: frontier open weights running on Arcen's own silicon, metered by the token and settled in USDG on Robinhood.
—
Tokens served
—
Inference calls
—
Models online
—
Uptime
The compute
One stack.
Zero markup.
We run the racks, the scheduler, and the rail. That means the posted token rate is the compute price. No gateway cut, no reseller cut, no per-call fee. The optimizations on the right are why we can afford to price it that way.
via the response cache
Continuous batching
Requests join a running batch mid-flight, so the GPUs never idle between calls, more tokens per GPU-second, cheaper tokens for everyone.
Prefix caching
Agents resend the same system prompt thousands of times a day. The cluster computes it once and reuses it, repeats become nearly free.
Quantized + speculative
Open weights served FP8-class with a draft model speculating ahead. Same output, materially more tokens per second per card.
Response cache
Identical repeat calls, retries, evals, loops, come back in ~10ms at half-price tokens. A posted billing rule, not a trick.
Fresh off the racks
The latest weights deployed to the cluster, every one priced at compute.
The token
Hold $ARCEN.
The cluster pays you.
Every call the network serves earns revenue on-chain. Half of it is split across $ARCEN holders every month, pro-rata to the supply you hold.
50%
of every call’s revenue fills the holder pool
Monthly
pro-rata payouts, straight to holding wallets
0
staking, lockups, or claim races
0x78a166a1b5968e4e31cd5728b8f4abc41cc2a175Drop-in
Your SDK already
speaks Arcen.
One email, one wallet
Sign in and a non-custodial wallet spins up in the background, no seed-phrase ceremony.
Load USDG, mint a key
Top up once and mint your ar_sk_ key. It's shown a single time, after that, it's yours to guard.
Swap one URL
Keep the OpenAI client you already ship. Point base_url at Arcen and every call meters and settles itself.
from openai import OpenAI
client = OpenAI(
base_url="https://api.arcen.com/v1",
api_key="ar_sk_live_…",
)
resp = client.chat.completions.create(
model="glm-5.2",
messages=[{"role": "user", "content": "Hello"}],
)
# metered + settled on-chain automaticallyYour agents are ready. Give them an engine.
Billing is exact: tokens used × posted rate, settled in USDG on Robinhood, no per-call fees, no gas, nothing recurring, nothing hidden. High-tier models always run pay-as-you-go.