What it does
Blackbox describes itself as a platform for running frontier models with a focus on trust and security. Teams can use hosted models from a catalog or deploy their own open weights, and reach them through a single API endpoint with one bill. The home page cites more than 300 models. A central feature is end-to-end encrypted inference: prompts are encrypted on the client and decrypted only where the model runs, and the site says zero data retention is enforced at the gateway. Dedicated deployment, a Microsoft Azure option and forward-deployed engineering are listed for larger customers. Enterprise customers can also rent NVIDIA B300 GPUs, configure the environment over SSH and add capacity as needed, with a stated 99.9% uptime SLA. For teams that want to adapt an open-weight model to their own data and tasks, the platform offers fine-tuning. The site also offers a model catalog, model comparison, a router, API reference and SDK guides. Several capabilities are labelled as available to Enterprise customers, and pricing is on a separate page that was not part of the text read.
Capabilities
- Model catalog and comparison behind a single API endpoint
- End-to-end encrypted inference with zero data retention
- Dedicated deployment on rented NVIDIA B300 GPUs
- Deploy your own open-weight models
- Fine-tuning of open-weight models on your own data
- Model router, API reference and SDK guides
When to use it
- Your organization needs private, encrypted inference for sensitive prompts
- You want one endpoint and one bill across many hosted models
- You want dedicated GPU capacity for open-weight models or fine-tuning
