SECURE INFERENCE FOR ENTERPRISE

Private,secureinferenceatrecordspeed.

Open frontier models at verified top speed, through one endpoint that your full stack can call. Zero data retention, at a much lower price than the closed labs.

454 t/s
NEMOTRON 3 ULTRA · #1 ON ARTIFICIAL ANALYSIS
0
DATA RETENTION · ENFORCED AT THE GATEWAY
2.7×
LOWER COST THAN THE #2 PROVIDER
300+
MODELS · ONE ENDPOINT, ONE BILL

5M+ developers, including engineers at the world's largest companies, build with BLACKBOX.AI

THE PLATFORM

One commit, two ways to run a model.

Enterprise Inference runs the open-weight model that you choose on our GPUs, isolated to you. The Blackbox Router connects you to the 300+ models that we do not host, with zero data retention and no training. One commit covers both products and the agent tools.

VIEW DOCS
01

Enterprise Inference

We deploy the open-weight model that you choose, only for you — reserved capacity, one tenant, verified top speed. No shared pools, and no other customers.

#1 Nemotron 3 Ultra provider · 454 t/s
Single-tenant · isolated to you
Customer-managed keys (BYOK/CMK)
02

Blackbox Router

One endpoint connects you to the 300+ models that we do not host. The gateway enforces zero data retention and no training. One key, one bill, one dashboard.

The gateway enforces zero retention
PII removed before closed models (Enterprise)
Encryption in transit
03

Agents & tooling

The Agents API, an AI-native IDE, the VS Code extension, and the CLI. Each surface uses the same per-token commit.

Works with every agent harness
SSO sign-in across every surface
One commit covers the full organization
BENCHMARKS

Independent measurements, not claims.

Artificial Analysis independently measured our output speed on NVIDIA Nemotron 3 Ultra. We are #1 of all providers — 30% faster than the #2 provider, at 2.7× lower price, with encryption in transit on.

tokens / second · nemotron 3 ultra · source: artificial analysis, july 2026
BLACKBOX.AI454
NEBIUS351
COREWEAVE222
TOGETHER AI173
SECURITY & PRIVACY

Your prompts are not our product.

Route across 300+ models and send only the data that each task requires. Run the workloads that you cannot share on a dedicated deployment that serves only you.

IN TRANSITWe encrypt all traffic in transit with TLS 1.3 and retain zero data
CLOSED MODELSOn Enterprise, we remove PII before prompts reach a closed model
TRAININGTraining opt-out is on by default, and we enforce it with the upstream providers that we route to
AUDITSSOC 2 Type II and ISO 27001 audits are in progress; HIPAA BAAs are available on request

We suppress retention and training on routed traffic through provider terms and per-request flags, wherever the provider API supports it.

READ THE SECURITY OVERVIEW
ISOLATION
ENTERPRISE APIOur endpoint, our GPUs
DEDICATED CAPACITYReserved for you
SINGLE-TENANT DEPLOYMENTIsolated to you
NO HUMAN REVIEWNegotiated per contract
The Enterprise API is our endpoint on our GPUs. We route each client to their own dedicated deployment, isolated from every other customer.
300+ MODELS · ONE ENDPOINT

Deploy privately.

MODEL
TYPE
CONTEXT
INPUT $/M
zaiglm-5.2
Text
1M
$1.40DEPLOY
moonshotaikimi-k2.7-code-highspeed
Code
262K
$1.90DEPLOY
moonshotaikimi-k2.7-code
Code
262K
$0.74DEPLOY
nvidianemotron-3-ultra-550b-a55b
Text
1M
$0.32DEPLOY
alibabaqwen3.7-plus
Text
1M
$0.32DEPLOY
minimaxminimax-m3
Text
1M
$0.60DEPLOY
VIEW THE FULL CATALOG

Open-weight models can run as a dedicated deployment on Enterprise — single-tenant, isolated to you.

Or start with three lines of code.

One OpenAI-compatible endpoint. Change the base URL and keep your code. You get intelligent routing across 300+ models, with encryption in transit on each connection.

OpenAI-compatible REST and streaming
Prompt caching and cost-aware routing included
Encryption in transit · zero data retention
Per-token Enterprise commits · rates improve with spend
infer.py
# pip install blackbox
from blackbox import Client

client = Client(api_key="fp-•••••••")

out = client.infer(
    model="nemotron-3-ultra",
    prompt="Explain inference at scale",
    stream=True,
)

for tok in out:
    print(tok.text, end="")
PROOF

Check it yourself.

BENCHMARKSIndependently measured speed#1 Nemotron 3 Ultra provider on Artificial Analysis. Open the live benchmark and verify the numbers.OPEN THE BENCHMARKPARTNERSHIPMicrosoft AI partnerLearn how the partnership works and what it includes for enterprise deployments on Azure.READ MORESECURITYThe security overviewEncryption, retention, deployment boundaries, and audit status — in plain language.READ MORE
High-trust machine intelligence, abundant and secure

Any model. Full speed.
Your data stays yours.