PRODUCT · ENTERPRISE INFERENCE

Your model. Your deployment. Yours alone.

You choose the open-weight model, and we run it as a deployment dedicated to you and isolated from every other customer, at independently verified top speed. One tenant, reserved capacity, and nothing shared — we operate it for you, as you would operate it yourself.

CAPABILITIES

The whole stack, isolated to you

Production inference on capacity reserved for one customer, with the controls that your security review will ask for.

Your deployment, your model, isolated

You select the open-weight model, and we run it as a deployment dedicated to you and isolated from every other customer. No shared pools, and no other customers — the capacity and the model instance are yours for the term of the contract.

Single-tenant, run for you

We operate it as you would operate it yourself: one tenant, one deployment, and no multiplexing across customers. Your forward-deployed engineer configures it with you at contract start.

Dedicated capacity

We reserve capacity and size it to your traffic, so your throughput does not change when another customer's job starts.

Nothing retained

We process prompts, code, and completions in memory and discard them after the response — nothing enters shared Blackbox storage. Abuse monitoring and human review can be disabled — negotiated into the contract, then enforced technically.

Your own weights

Upload fine-tuned models to a private registry and call them through the same endpoint, on the same reserved capacity.

OpenAI-compatible API

Point your existing SDK at your endpoint. The request and response shapes are the same — no rewrites, and no new client.

Identity and audit

SAML SSO, SCIM provisioning, fine-grained RBAC, and tamper-evident audit logs streamed to your SIEM.

HOW IT WORKS

Tokens, start to finish

From deployment to streamed token: the platform manages scale and capacity, and the tenancy stays only yours.

  1. 01
    We create your deployment

    A forward-deployed engineer configures a single-tenant deployment of the model that you chose, dedicated to your account and isolated from every other customer.

  2. 02
    Your request stays inside it

    Open-weight models run on capacity reserved for you. We forward nothing to a third-party provider unless you deliberately route it there through the Router.

  3. 03
    Tokens stream back

    Tokens stream as the model generates them, encrypted in transit. We process them in memory and discard them after the response.

QUICKSTART

Streaming in five lines

Point the OpenAI SDK at your endpoint and stream tokens — the rest of your code is unchanged.

TYPESCRIPT
import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.BLACKBOX_API_KEY,
  baseURL: "https://api.blackbox.ai/v1",
});

const stream = await client.chat.completions.create({
  model: "nemotron-3-ultra",
  messages: [{ role: "user", content: "Stream me a haiku about GPUs." }],
  stream: true,
});

for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta?.content ?? "");
}
PERFORMANCE

Independent measurements, not claims

Artificial Analysis independently benchmarked NVIDIA Nemotron 3 Ultra, and measured latency with the same methodology.

454 t/s
NEMOTRON 3 ULTRA · #1 ON ARTIFICIAL ANALYSIS
+47%
FASTER THAN THE #2 PROVIDER
$0.44
BLENDED / 1M · 2.7× CHEAPER THAN #2
6.1 s
TIME TO FIRST ANSWER TOKEN (AA)
FAQ

Questions, answered

Is it compatible without code changes?

Yes. Change the baseURL and apiKey on the official openai SDK, and existing requests, parsing, and tooling keep working. Streaming uses the same stream: true flag.

What happens when traffic increases quickly?

Endpoints autoscale across the fleet and route to warm GPU capacity. Bursts return 503 only while capacity scales. Retry with backoff, and the request completes.

Can I serve my own model?

Yes. Enterprise includes a private model registry. Upload your weights and call them through the same endpoint, on the same reserved capacity.

Who runs the deployment?

We do. Blackbox operates it on our own infrastructure as a single-tenant deployment dedicated to your account — as you would run it yourself, but we run it for you. It is isolated from every other customer: your own model instance, your own reserved capacity, and no shared pools. You do not install or operate it in your own infrastructure — we run it end to end.

What about the models that you do not host?

You reach those through the Blackbox Router — a separate product, with a deliberately separate promise. The gateway enforces zero data retention and no training through provider terms and per-request retention flags, wherever the provider API supports it. But the model still runs on the infrastructure of a third party, and the gateway must read the request to route it. Isolation is what this product adds: one tenant, one deployment, and nothing forwarded anywhere. Workloads that cannot accept a third-party provider belong here instead.

Inference with one tenant only.

Tell us the workload and the controls that it must satisfy. We reply with a deployment plan and a per-token commit.