DeepInfra
Verified May 27, 2026DeepInfra runs open-weights models at some of the lowest per-token rates on the market — typically 20–40% below the median for the same Llama, Qwen, or DeepSeek model. The catalog is broad but pricing is the headline feature.
Serverless endpoints cover most chat models, while dedicated GPU instances are billed per-second for users who need predictable capacity. Strengths are cost and uptime on common models; weaknesses are slower onboarding for brand-new releases and a thinner story on enterprise compliance versus Together or Fireworks.
Inference is served from US data centers with limited regional options. For teams optimizing on dollar-per-million-tokens without strict residency requirements, DeepInfra is usually the cheapest credible option among the open-weights hosts.
Calculate cost for your workload
Plug in your monthly tokens — get the actual bill on every provider serving each model.
Open calculator