Data you can inspect.
The measurements are useful only if you can see what produced them.
TL;DR
Start with the catalogue, then open a report’s evidence links. The media dataset includes every retained output and the recorded failures. Older reports have uneven coverage; missing files are labelled instead of reconstructed or invented.
Download the records
- Complete catalogue · JSON — titles, dates, configuration metadata and links to available evidence.
- Browse all 145 records — selected reports and historical documents.
- October media campaign · full dataset — prompts, timings, outputs and measurement limits.
- Media data guide and attribution.
- Restored-download publication manifest — file hashes and the privacy redactions used for restored raw downloads.
Source and the older archive
The complete editable website lives in MyServers/web/benchmarks/ on the private Gitea repository used to maintain this site. The older public benchmark archive on GitHub contains published run bundles and a contribution harness. It is a historical archive and may not contain the latest website changes.
Find your hardware
Pavilion · GTX 1050 4 GB · Predator · GTX 1060 6 GB · Mac · Historical CPU-server records
License and citation
Benchmark data is published under CC BY 4.0; helper code uses MIT where stated. Generated media and third-party models can carry separate terms: check the campaign attribution and model license before reusing them.
Cite the individual report URL, its recorded test date and the configuration you used. For example: “Margetic, S. (2026). Weeyuga Benchmarks. [report title and URL], accessed [date].”