Provider pricing

Lepton AI AI API pricing

Lepton AI offers serverless inference for popular open models (Llama 4, DeepSeek, Qwen, Mixtral, Stable Diffusion) plus dedicated GPU deployments. Optimized for sub-second cold starts and high throughput, with per-token billing on serverless and per-GPU-hour on dedicated. Acquired by NVIDIA in 2025.

Founded2023
HeadquartersPalo Alto, CA
Tracked modelsDirectory

Best for

Teams that need predictable low-latency inference on open models without managing infrastructure.

Strengths

  • Sub-second cold starts on serverless endpoints
  • Dedicated GPU deployments with autoscaling and SLA
  • Photon framework for one-click custom-model deployment
  • NVIDIA-backed roadmap and reserved GPU capacity