How we check prices

Last run: 21 Oct 2026, 10:07 am AEST

When

A script runs every morning at about 4am Brisbane time (AEST). It visits each store, records every matching listing with its price, stock status, link and the time it was read, then rebuilds the site. If a run fails, the site keeps the last good data and the affected prices are marked.

Which stores

StorePages read last runStatus
Mwave
1
problem on last run
Scorptec
1
ok
Umart
1
ok
Apple AU
2
ok
Minisforum AU
4
ok
GMKtec (USD)
2
ok
Framework
0
problem on last run
Microsoft Store AU
0
problem on last run
PC Case Gear, Centre Com, JW Computers
0
block automated checks, so not included
PLE
0
prices load in JavaScript, not included yet
Amazon AU
0
needs Amazon's product API, not included yet

We read public category pages and product feeds only, follow each store's robots.txt and crawl delay, and skip a store that shows a bot check. US-dollar prices are never converted and never used as the "price today"; they show up as "reference only" where we have them.

Matching listings

Each device has a short list of name rules (for example "RTX 5090" and "32GB"). Listings with words like "laptop", "bundle", "enclosure" or "gaming PC" are dropped, so bundles and pre-built PCs don't count as the card's price. Apple prices are Apple AU's "From" price for the named configuration.

In stock

A listing counts as in stock when the store's own data marks it as available: the product feed's "available" flag, the page's schema.org InStock tag, or the store's stock indicator. If a store doesn't say either way we show "stock unknown". Some stores flag pre-orders as available, so if you spot one we've called in stock, please report it below. The headline price is the cheapest in-stock listing. If a cheaper listing exists that isn't in stock, the page says so.

Price history

We keep the lowest price of each day per device. Charts and "up/down since" figures come from that log, which started on 19 Oct 2026, so short histories are labelled as such.

Model fit and speed

"Fits" means the model's weights, its context cache at 8k tokens and 1GB of overhead come to 90% or less of the device's usable memory. For graphics cards, usable is VRAM minus 0.5GB. For Macs and unified-memory PCs we count 75%, the default GPU limit. Speeds are measured where a public llama.cpp benchmark exists for that device and estimated from memory bandwidth otherwise, and they're labelled either way. Full formulas are on the methodology page.

What we can't see

Corrections

If something's wrong, we fix the rule or the data, rebuild, and the fix shows on the next run (usually the next morning). We don't edit individual prices by hand to make them look better.