Run any AI model for less.
Find the cheapest way to run the model you need — a hosted API, a rented GPU, or a whole node. Live prices from GPU clouds and AI API providers, matched to the models they can run.
Popular models, and what they cost to run.
All AI models →For each model: the cheapest hosted API, and the cheapest single GPU that holds it at 4-bit — both at today's prices.
Pick a GPU by what you need to fit.
Check a specific model →Memory decides which models a GPU can run; price decides what you pay. Here are the cheapest cards in each memory class right now.
7–8B models at full precision, 14B at 8-bit, ~32B at 4-bit. Image generation.
70B models at 4-bit, 32B at 8-bit. The value tier for private LLMs.
70B at 8-bit, ~120B at 4-bit, QLoRA fine-tuning of 70B models.
70B at full precision on one card; the building block for DeepSeek-class MoE models.
Who runs AI for you.
GPU clouds & marketplaces
See all →Rent a GPU by the hour and run anything on it.
Hosted AI APIs
See all →Pay per token for models someone else serves.
What are you building?
Latest analysis.
All articles →Every price here comes from the providers' own listings — GPU marketplaces every few minutes, fixed-price clouds and API catalogs daily. Rankings are by price only; affiliate links never change the order.