DeepInfra

★ 4/5 · Inference Platform

DeepInfra is a low-cost inference API that lets developers run open-source LLMs, embedding and image models on managed GPUs with simple per-token pricing.

Pros

Lowest-cost LLM inference starting at $0.06/million tokens (Llama 3.2 3B)
serverless pay-per-use model requires no infrastructure management
batch API processes 1,000 requests at 50% discount for non-urgent workloads

Cons

Limited to open-weight models (no proprietary models like GPT-4)
pricing varies wildly by model ($0.06-$82.50/M tokens) making budget forecasting difficult
smaller model library compared to OpenAI or Anthropic

Cost

Pay-per-use: from $0.03-$0.06/M tokens (small models) up to ~$0.60/M tokens (large models, e.g. Llama 4 Maverick output); Llama 3.3 70B at $0.35 input + $0.40 output per million tokens; batch API discounts available

Verdict

Best for cost-sensitive teams running open-source models at scale, but unsuitable for applications requiring closed-source LLMs.

Visit DeepInfra