Which local AI models can your hardware run?

Hardware profile

Tell us what data has to move through.

Choose the closest specification. You can overwrite GPU memory where a variant differs.

Processor
System memory
Graphics
Motherboard connection

Workload requirements

Required capabilities
Working context How much should one session retain?
32K
Long documents · approximately 24k words~0.9 GB KV cache for an 8B Q4 model
4K8K32K64K128K256K
Capacity policy Control what the calculator must leave untouched

Local storage disclosure: Your configuration is stored only in this browser so it can be restored on return. It is not submitted to Paradox Design.

Read the result as a shortlist.

The calculator narrows model options from the CPU, RAM, GPU and connection you enter. A model fitting in memory does not establish acceptable speed or answer quality. Context length, quantisation and simultaneous users all change workload demand.

Memory

Check model weights and working memory alongside the operating system and other services. GPU VRAM and system RAM are not interchangeable for performance.

Speed

Test the time to first response and sustained output with representative prompts. A shared service also needs a concurrency check.

Quality

Compare candidate models on approved examples from the actual task. A smaller model that answers the task reliably may be the better fit.

External project website

Open this project site?

is hosted at . It will open in a new tab.