Blackbox

Overview

Blackbox is a high-trust platform for frontier inference. It runs the open-weight model you choose as a dedicated deployment on Blackbox GPUs, or routes 300+ models through one endpoint. Prompts are encrypted before they leave your machine and decrypted only where the model runs, so nothing readable crosses the wire.

The gateway enforces zero data retention and keeps training opt-out on by default. Artificial Analysis independently measured Blackbox as the #1 Nemotron 3 Ultra provider, at 454 tokens per second and 2.7× lower cost than the #2 provider. Route across 300+ models and send only the data each task requires, or run workloads you cannot share on a deployment that serves only you.

How It Works

You commit to a token volume through a purchase order, and the balance decreases as your teams consume it at published per-model rates. A forward-deployed engineer then configures the account, including controls such as PII removal, at contract start. For API access, you point three lines of code at an OpenAI-compatible endpoint and keep your existing code.

Why Choose This Product

Blackbox operates its own GPUs end to end for open-weight models, which is why open-weight commits earn a larger discount than closed models. Closed models route to their upstream providers and bill through those relationships. There are no platform fees, no markups, and no per-seat charges, and metering runs per token rather than per seat.