Real jobs, not lab numbers
Every laptop runs the same 6 everyday business jobs: contract summary, customer replies, marketing pack, invoice extraction, sales data questions, handbook q&a. A verdict for each job tells you what it is like to use.
Hardware for running AI on your own devices
We run 6 everyday business jobs on each laptop with AI that stays on the machine. Each laptop gets a plain verdict on every job and one score out of 100. Your documents never leave your computer.
Not sure what your laptop can run? Try the finder. Estimates are labelled.
Summarise this contract and list the key dates.
Write a polite reply. Offer a refund.
Pull out the supplier, invoice number and date.
Every laptop runs the same 6 everyday business jobs: contract summary, customer replies, marketing pack, invoice extraction, sales data questions, handbook q&a. A verdict for each job tells you what it is like to use.
The AI runs on the laptop itself. Our tests use made-up documents, and the setups we build keep your files on your own machine.
The test rules are written down and locked before we test. Every score is free to download. Read the method or get the data.
Hardwarerig is about running AI on your own computer, so your files stay with you. Here is what you can use on the site now.
The model is installed on the machine, so your file goes in and the answer comes out without leaving it. We choose the laptop, set it up and test it on real jobs.
$ ollama run qwen3:8b "Summarise this contract and list the dates" < contract.txtTwo parties agree a 12-month service, paid monthly.It starts on 1 March. Either side can end it with 30 days' notice.A fee applies if a payment is more than 14 days late.$
The picture on the left shows an open model running through Ollama, a free tool that runs AI models on your own computer. You do not need to use a terminal: we set up an app on the laptop and show you how to use it.
Each scorecard shows a verdict for every job, a score out of 100 and the laptop's specifications. We publish one only after a full test.
Pick the job. We show how laptops do at it.
summarise long contracts
answer customer emails
write marketing copy
pull data out of invoices
answer questions about sales data
search long internal documents
The finder works from the specifications that makers publish and from benchmark results that others have published, so it covers laptops we have not tested ourselves. Pick a laptop, a chip or a graphics card and see which open AI models fit in its memory and roughly how fast they write.
Every result is labelled Estimated from specs or Measured, and laptop pages show Reported by others benchmarks where they exist. Speeds are ranges. Measured means we ran the test ourselves.
Renting a cloud GPU needs no hardware purchase and adds up by the hour. Your own machine costs more up front and less after. Enter your own numbers and see where the two cross.
Cloud rates come from the public price pages of Crusoe, DigitalOcean, Hyperstack, Lambda, RunPod, Thunder Compute, Vast.ai, Verda: 151 GPU and provider listings, each dated and sourced. The latest were read on 2026-10-28.
6 everyday business jobs, with the same prompts and the same documents each time.
The AI model runs on the machine being tested, with the settings locked in advance, so the result is what you would get.
Each job gets a verdict and each laptop one score out of 100. We publish a scorecard only after a full test and never publish estimates as scores.
We can install private AI on your laptop and show you how to use it. Every job is quoted for the client.
This one is still being built, one industry at a time. See the industries.
Local AI keeps your files on your own machine, so they are never sent to a cloud service.
We are starting with one industry at a time. Join the list for yours.