Run open-source LLMs, of any size.

Paste any Hugging Face link, and we'll automatically download the weights and boot a machine with the correct number of GPUs.

You can use up to 640GB VRAM, with simple, pay-as-you-go pricing.

Broad support.

We run models and finetunes of any supported architecture. Supported architectures include:

• Meta Llama 3.x
• Mistral

  (and Mixtral)
• Qwen 2.5
• Phi-4

And many more. If vLLM supports it, we do too.

💫 New: we also support DeepSeek V3 by proxying to Together.ai!

Get started: