LLM rental with hourly GPU billing

Run open-weight models on dedicated Qudata GPUs instead of paying per token.

Pick a model and we bring it up on a matching GPU behind an OpenAI-compatible API. Prompts and data never leave the rented server, there is no cap on requests, and you pay only for GPU hours.

Qwen

By Alibaba Qwen

Single-GPU flagship

Runs on 1x RTX5090~42 tok/s

Deploy

Gemma

By Google

Largest Gemma on one GPU

Runs on 1x RTX5090~38 tok/s

Deploy

GPT-OSS

By OpenAI

Open reasoning model for agents

Runs on 1x RTX3090~27 tok/s

Deploy

GLM

By Z.ai

Fast GLM for code and agents

Runs on 1x RTX5090~52 tok/s

Deploy

Llama

By Meta

Meta's model for local agents

Runs on 1x RTX5090~38 tok/s

Deploy

NVIDIA

By NVIDIA

NVIDIA's multimodal flagship

Runs on 1x H200~35 tok/s

Deploy

Poolside

By Poolside

Local coding agent

Runs on 1x RTX5090~48 tok/s

Deploy

Cohere

By Cohere

Cohere's coding MoE

Runs on 1x RTX5090~52 tok/s

Deploy

Mistral

By Mistral AI

Strong compact Mistral

Runs on 1x RTX3090~36 tok/s

Deploy

Kimi

By Moonshot AI

Flagship MoE for agentic coding

Runs on 3x H200~19 tok/s

Deploy

DeepSeek

By DeepSeek

Hybrid DeepSeek flagship

Runs on 3x H200~22 tok/s

Deploy

IBM

By IBM

IBM enterprise all-rounder

Runs on 1x RTX5090~38 tok/s

Deploy

Phi

By Microsoft

Enhanced reasoning model

Runs on 1x RTX3090~36 tok/s

Deploy