CanRun

Terms of use

Last updated October 25, 2026.

These terms explain what CanRun is, what its numbers can and can't tell you, and how you can use what you find here. They apply whenever you use the site. If anything is unclear, ask us through the contact page.

What CanRun is

CanRun is a free website that estimates which open large language models a graphics card, Mac or PC can run locally, which quant to download, how much memory each one needs and roughly how fast it will generate text. There are no accounts and nothing to buy on the site, and the calculator runs in your browser.

Results are estimates

  • Every verdict, speed and memory figure is an estimate for reference. They are worked out from published specs, real file sizes and public benchmarks, and they can be off.
  • Your own results depend on things CanRun can't see: driver and software versions, runtime settings, the context length you actually use, other programs taking up memory, and cooling or power limits.
  • That is why speeds are shown as ranges with a confidence label. How the numbers are made, how far they have been off and where they fall short are on the How we calculate page.
  • The data can be out of date or wrong. The date of the current data is at the bottom of every page. If a number looks wrong, please tell us.

Your decisions are your own

CanRun can help you compare hardware and models, but nothing on it is advice to buy, sell or invest in anything. Before buying a graphics card or a computer, check the specs and price with the seller and, if you can, results from people who own that hardware. Decisions you make based on CanRun's estimates, including purchases, are yours.

Outside links and model licenses

  • CanRun links to sites it doesn't run, such as Hugging Face, Ollama, llama.cpp and hardware makers. Those sites have their own terms and privacy policies, and we can't vouch for their content or for how it changes.
  • Open models come with their makers' licenses. Some limit commercial use or set other conditions. CanRun shows a short license label for each model, but the license on the model's own page is the one that applies, so read it before you use the model.
  • Product names and trademarks belong to their owners. CanRun is not affiliated with any GPU maker or model publisher.

Ads and affiliate links

CanRun may show Google AdSense ads, and may later add affiliate links to stores, which pay us a small commission when you buy through them at no extra cost to you. There are no ads or affiliate links on the site right now. If they are added, they will not change verdicts or the order of results. How advertising cookies work is explained in the Advertising section of our privacy policy, which we will update before anything goes live.

Quoting and reusing our content

  • You are welcome to quote CanRun's numbers or short passages, for example in a forum post, an article or a video. Please name CanRun, link to the page you quoted and mention that the numbers are estimates.
  • Please ask first before copying whole pages or republishing the data in bulk, for example by scraping the site.
  • The source code and the raw results of our own benchmarks are public in the sejong00/canrun repository on GitHub, so anyone can check how the numbers are made. The repository has no open-source license at the moment, so if you want to reuse the code or data beyond quoting it, please ask.
  • Data that comes from other sources, such as model details, hardware specs and public benchmarks, stays subject to those sources' terms. The Sources section of How we calculate lists them.

Changes to the site

CanRun is provided as it is. Pages, models and hardware can be added or removed, the way estimates are calculated can change, and the site may sometimes be unavailable. Changes you would notice are listed on the Updates page.

Changes to these terms

If we change these terms, we will update this page and the date at the top.

Contact

Questions about these terms: [email protected]. Other ways to reach us are on the contact page.