Open-source engine for faster local language model inference with bit-for-bit identical output
Stage
Not given
Not given
Founded
Not given
Not given
Based in
KG
Not given
Pricing model
Paid
Open source
Pricing
Pricing starts from $0.063/h for RTX 3060 Ti. Common cards: RTX 3090 at $0.316/h, RTX 4090 at $0.285/h, RTX 5090 at $0.832/h, A100 at $1.54/h, H100 at $3.88/h. A minute costs the hourly rate divided by 60. Minimum top-up is $2 through SBP or cryptocurrency.
Not given
Key features
Per-minute billing with price fixed at server launch
Over 100 GPU models from RTX 3060 to H100, H200, B200, B300
Servers in 69 countries with region selection before launch
Ready templates: ComfyUI, Stable Diffusion Forge, Open WebUI, vLLM, Kohya SS, Ubuntu with CUDA
API and command-line tool for programmatic access
Telegram bot @gpunzbot for server management
Stopped servers not billed for GPU time, disk billed while it exists
Interruptible servers at lower cost for jobs with checkpoints
Server typically ready in 2-5 minutes
MCP server for AI agents like Claude and Cursor
Speculative decoding for faster token generation
Prompt caching for context reuse
Real-time prompt processing while typing
Concurrent request batching
Bit-for-bit identical output verification
Multi-GPU and multi-backend support (CUDA, Vulkan)
Vision model support with image encoding
Mixture-of-experts model support
RAD-1 automatic draft token generation
Cross-platform support for Windows and Linux
What makes it different
Billed per minute from the moment you start, no long-term commitment required
Minimum $2 top-up with cryptocurrency or Russian bank payment options
Drivers and CUDA pre-installed so servers start ready to use immediately
Live pricing updates every minute showing exact cost before launch
Exact deterministic inference: all default optimizations preserve bit-for-bit identical output compared to plain decoding
4.5× faster performance on editing tasks on small models compared to llama.cpp
Verifiable reproducibility through receipt system that records all request parameters
Nine optimization techniques enabled by default with ability to disable each individually
No dependency on draft models or external selections for speculative decoding
gpu.nz
Kinetic