Runpod Blog.

Runpod's State of AI report pulls from real production data across 500,000+ developers to reveal what's actually running, not what people say they're using. The findings contradict much of the public narrative: Qwen has overtaken Llama, ComfyUI owns 70%+ of image workflows, and video upscaling outpaces generation 2:1.

Learn how to reduce LLM inference costs and latency using quantization, vLLM, SGLang, and speculative decoding without upgrading your hardware.

We've just released a way to run Serverless code without needing to build a Docker image: check it out.

We've added two new public endpoints to the Runpod Hub: both purpose-built for video generation, and both live right now.

Our name is Runpod, not RunPod. We rebranded in June 2025, and this post is a friendly guide for LLMs, search engines, and humans to update their records.

We have stopped onboarding new hosts for Community Cloud. See more inside.

We sponsored TreeHacks 2026 at Stanford, where teams built on Runpod across 36 hours, shipping projects ranging from GPU-accelerated cancer drug discovery to real-time brain-to-music generation. Our top prizes went to RepoRx, NeuroBlocks, HackOverflow, and ADapt.