Deep Infra

What's good about Deep Infra

• Simple OpenAI-compatible API alongside controllable GPU rentals
• Competitive hourly rates for NVIDIA GPUs including B200 and B300
• Fast provisioning with SSH access for dedicated instances (ready in ~10 seconds)
• Supports custom deployments in addition to hosted public models

Deep Infra GPU prices

5 GPU models across 5 configurations. Rank compares Deep Infra's cheapest on-demand listing for each GPU with every provider's cheapest.

What a run costs on Deep Infra

24 hours at today's cheapest listing for each GPU, at the size it is listed.

• 1× H100 SXM for 24 hours

  At $2.90/GPU-hr

  $69.60

  $7.08 under the market median rate

• 1× H200 for 24 hours

  At $3.49/GPU-hr

  $83.76

  $25.19 under the market median rate

• 1× A100 SXM for 24 hours

  At $1.49/GPU-hr

  $35.76

  $1.32 under the market median rate

The market median is every provider's cheapest on-demand price for that GPU.

Deep Infra LLM API prices

Pay-per-token pricing, per 1M tokens.

Prices last updated November 1, 2026

Billing & terms

Serverless pay-per-token

OpenAI-compatible inference APIs billed per input and output token, with no idle GPU time or minimums

Dedicated GPU hourly rates

Published transparent hourly pricing for A100, H100, H200, B200, and B300 GPUs with pay-as-you-go billing

Per-execution-time billing

Non-LLM models are billed for inference execution time rather than per token

Cached input pricing

Prompt-cached input tokens are billed at a lower rate than uncached input tokens

No long-term commitments

Flexible hourly billing for dedicated instances with no prepayments, contracts, or minimums required

Regions

Runs on self-operated, inference-optimized infrastructure in US-based data centers; no international region list is published.

Support

Documentation site, dashboard guidance, Discord community, feedback email (feedback@deepinfra.com), and contact-sales options.

Watch out for

• Infrastructure is US-based only, with no published international regions
• Primarily focused on inference and GPU rentals rather than broader cloud services
• Newer player compared to established cloud providers

Features

Drop-in OpenAI Replacement

OpenAI-compatible API for 100+ models including DeepSeek, Qwen, Llama 4, Claude, and Gemini families with autoscaling

Dedicated GPU Rentals

B200 instances with SSH access spin up in about 10 seconds and bill hourly

Custom LLM Deployments

Deploy your own Hugging Face models onto dedicated A100, H100, H200, B200, or B300 GPUs

Transparent GPU Pricing

Published per-GPU hourly rates for A100, H100, H200, B200, and B300 with competitive pricing

Owned Inference Infrastructure

Runs on its own inference-optimized hardware in US-based data centers rather than rented capacity

Comprehensive AI APIs

Support for text generation, vision and OCR, embeddings and reranking, image, video and music generation, and speech recognition and synthesis

Batch API and Prompt Caching

Asynchronous batch endpoints with a files API, plus prompt caching billed at a reduced cached-input rate

Framework Integrations

Documented integrations for LangChain, LlamaIndex, the Vercel AI SDK, AutoGen, and the Anthropic SDK including Claude Code

SOC 2 & ISO 27001 Compliance

Zero retention policy with enterprise-grade security certifications for data privacy and protection

Services

• Serverless Inference

  Hosted model APIs with autoscaling on Deep Infra's own inference infrastructure.

  • Drop-in OpenAI replacement - swap base URL, keep existing code
  • 100+ models including the DeepSeek V4, Qwen 3.7, Llama 4, Claude 5, Gemini 3.5, Gemma 4, GLM-5, Kimi K2.7, and Nemotron 3 families
  • Comprehensive AI capabilities: text, vision, embeddings, image/video generation, speech
  • Always-fresh model catalog with new releases deployed quickly
  • Per-model service tiers (priority and flex) and a batch API for asynchronous jobs

• Dedicated GPU Instances

  On-demand GPU nodes with SSH access for custom workloads.

• GPU Clusters

  Multi-node B200 and B300 clusters with SSH access for training and full-control workloads.

  • Rent a B200 or B300 cluster and run arbitrary workloads
  • On-demand DGX B300 configurations billed per instance-hour
  • Sold alongside dedicated instances via the pricing page and contact-sales flow

• Private Model Deployments

  Run your own fine-tuned weights on dedicated hardware behind a private endpoint.

  • Deploy custom LLMs and LoRA adapters on A100, H100, H200, B200, or B300
  • Autoscaling with a private endpoint for data isolation and compliance

Getting started

1. Step 1

   Create an account

   Sign up (GitHub-supported) and open the Deep Infra dashboard

2. Step 2

   Enable billing

   Add a payment method to unlock GPU rentals and API usage

3. Step 3

   Pick a GPU option

   Choose serverless APIs or dedicated A100, H100, H200, B200, or B300 instances

4. Step 4

   Launch and connect

   Start instances with SSH access or call the OpenAI-compatible API endpoints

5. Step 5

   Monitor usage

   Track spend and instance status from the dashboard and shut down when idle