RunPod
Rent GPUs by the second for training, fine-tuning, and serverless inference.
Visit RunPod Pay per GPU-second; entry GPUs from under $0.50/hr
Quick verdict
- Best for
- Fine-tuning and experiments on a budget
- Pricing
- Pay per GPU-second; entry GPUs from under $0.50/hr
- Not ideal if
- Availability of popular GPUs fluctuates
What is RunPod?
RunPod is a GPU cloud that offers on-demand and spot GPU pods, serverless endpoints that scale to zero, and templates for popular models, at prices well below the big cloud providers. Popular with indie developers and startups training or serving their own models.
Key features
- On-demand and spot GPU pods from consumer to H100-class
- Serverless endpoints with scale-to-zero
- One-click templates for Stable Diffusion, vLLM, Ollama, and more
- Network volumes for persistent storage
- Per-second billing
Pros
- Among the cheapest GPU access available
- Serverless makes small inference services cheap to run
- Templates get you from zero to a running model in minutes
Cons
- Availability of popular GPUs fluctuates
- Community cloud tier has variable reliability
- You manage the software stack yourself
Best for
Fine-tuning and experiments on a budgetHosting open models without a platform markupBursty inference workloads
Read more
Related comparisons
Alternatives to RunPod
Modal
Serverless GPUs for AI — deploy any Python function at scale, pay per second.
Freemium$30/mo free credits, then pay-per-second
Released October 2021
Baseten
Production inference infrastructure for deploying your own AI models.
PaidUsage-based GPU pricing; enterprise plans
Released January 2019
Together AI
Fastest inference for open-source models — Llama 4, Qwen3, DeepSeek V3 at low cost.
FreemiumFree credits, then $0.20-$5/M tokens
Released June 2022