What is Inference?

Inference.net is a fully managed inference infrastructure platform for AI-native teams. It lets you deploy open-source, custom, and fine-tuned models on dedicated high-performance infrastructure, monitor production LLM traffic across any provider through a unified gateway, evaluate model quality, cost, and latency, and fine-tune specialized models on your own data. The platform is designed to serve frontier-quality models at scale with 99.99% uptime and transparent pricing.

How to use Inference?

1. Talk to an engineer or get started from the website. 2. Install the SDK and point your LLM traffic at the Inference.net gateway to capture traces, latency, and cost data. 3. Deploy models from the catalog or bring your own weights on dedicated infrastructure. 4. Use Observe and Evaluate to monitor quality, latency, and spend across providers. 5. Fine-tune custom models with your production data and push them directly into deployment.

Core features of Inference

Deploy: fully managed, global inference infrastructure with 99.99% uptime and per-hour dedicated GPU pricing
Observe: LLM observability and tracing across any model or provider via a single gateway
Evaluate: continuous benchmarking of model quality, latency, and cost against production traces
Train: automatic fine-tuning workflows for task-specific, frontier-quality custom models
Trace: end-to-end capture of prompts, tool calls, responses, and downstream provider behavior
HALO: open-source agent optimization that analyzes traces and ranks failure modes
Model catalog: deploy open-source models such as Kimi, MiniMax, GLM, and GPT-OSS
SOC 2 Type II compliant with full control over data and model weights