Rent exactly as much GPU as you need.
A slice of a GPU or the whole card, by the minute. SSH, JupyterLab and VS Code on every GPU, shared volumes, and an OpenAI-compatible inference API billed per token. Prepaid, no contracts.
What you get
A real GPU, sized to the job
Every GPU is a container on a real NVIDIA card, with root access and its own CPU, RAM and disk.
Slices with hard limits
Take a quarter, a half or the whole card. Memory limits are enforced, and each slice is guaranteed its share of the card's compute.
SSH, JupyterLab, VS Code
Root over SSH with a password or your key, a browser terminal, JupyterLab, and Remote-SSH for VS Code or Cursor. Ubuntu with CUDA and PyTorch.
ssh -p 30568 [email protected]Storage that outlives the GPU
Each GPU has a /workspace disk that survives pause and resume. Shared volumes belong to your team, mount on several GPUs at once and stay after you terminate.
Shared volumes US$0.10 / GB-month
Inference API
Point any OpenAI SDK at our base URL and use an API key. Chat, completions and embeddings on open models, billed per token.
curl https://flint.monster/api/v1/models -H "Authorization: Bearer $API_KEY"Prepaid, by the minute
Top up from US$25.00. A running GPU is billed per minute from your balance. Paused costs 25% of the hourly price; terminated costs nothing.
Spot for work that can wait
Spot slices cost about half. When on-demand neighbours on the card are busy, a spot slice's GPU work pauses and then carries on; your process and memory are kept.
How it works
From balance to shell in three steps
- 01
Add credit
Top up your balance by card, from US$25.00. Turn on auto top-up if you like.
- 02
Pick a GPU and launch
Choose a slice or a whole card. Launching sets aside 30 minutes of credit, and you pay only for the minutes you use.
- 03
Connect and work
SSH, the browser terminal, JupyterLab or VS Code. Pause to keep it for later, terminate when you're done.
Start with a slice. Scale to the card.
No contracts and no monthly minimum. Pay for what you run.