for hosts · rates and model compatibility

What your GPU earns

Search for your card and pick your region to see what you'd be paid and which AI models it can serve.

RTX 4090 · 24 GB
$73 a month
$0.306 an hour · $2.45 a day, rented 8 hours a day
16/22
models it can run

Models it can run

full contextshorter context, to fit in memory
  • Qwen3 0.6B16-bitup to 40K context
  • Gemma 3 1B16-bitup to 32K context
  • Qwen2.5 3B4-bitup to 32K context
  • Mistral 7B v0.34-bitup to 32K context
  • Llama 3.1 8B4-bitup to 128K context
  • Qwen2.5 7B4-bitup to 32K context
  • Qwen3 8B4-bitup to 40K context
  • Gemma 3 12B4-bitup to 128K context
  • Qwen3 14B4-bitup to 40K context
  • GPT-OSS 20B4-bitup to 128K context
  • Mistral 7B v0.216-bitup to 32K context
  • DeepSeek Coder 7B16-bitup to 4K context
  • Llama 3.1 8B16-bitup to 59K context
  • Qwen2.5 32B4-bitup to 17K context
  • Qwen2.5 Coder 32B4-bitup to 17K context
  • DeepSeek R1 32B4-bitup to 17K context

Context lengths assume the card is otherwise idle. The engine measures what actually fits when it starts, so a card doing other work serves less.

List your GPU

How you're paid

Per rented hour
You earn for the hours your GPU is serving someone, not for time online and idle.
Rate set at sign-up
It depends on your card and region, and is fixed when you register the GPU.
Every 15 days
Paid to your bank through Stripe once your balance reaches $10.
+$5 per payout
For early suppliers, on every payout during their first 3 months.

Monthly figures are estimates: what you make depends on how often your GPU is rented.

Why some models don't fit

A model has to fit in your card's memory. With room to spare it gets its full context — how much text it can read at once; a tight fit still runs, with a shorter one. 4-bit versions take about a quarter of the memory of 16-bit ones, which is how bigger models fit on everyday cards. RTX 20-series GPUs can't run some of these formats at all.