DeepInfra

DeepInfra offers low-cost serverless inference for open-source LLMs with pay-per-token billing, latency tiers, discounted batch inference, and dedicated GPU rentals.

DeepInfra provides serverless inference for open-source large language models and other AI models through a simple, OpenAI-compatible API. Developers pay only for the tokens they use, with no minimums and no charge for idle time, making it a cost-efficient way to run models like Llama, Mistral, DeepSeek, and many others in production without managing GPUs. Pricing is organized into latency tiers: Flex bills at 0.8x the base per-token rate for non-production and asynchronous work, the default Standard tier bills at 1x, and Priority bills at 1.5x for faster time-to-first-token during peak demand. DeepInfra also offers serverless batch inference at roughly half price with results delivered within 24 hours, and dedicated GPU-hour rentals from A100 up to B300 for teams that want reserved capacity. Per-token rates span a wide range depending on model size. DeepInfra raised a $107M Series B in May 2026, reflecting strong growth in the serverless inference market. It suits developers and teams that want the lowest-cost path to running open-source models at scale, competing with providers like Together AI, Fireworks, and Groq on price and model coverage.