Corrections and change log
When a number on RunsOnMy is confirmed wrong, the fix goes into the shared code and is listed here with the date, what was wrong and what changed. Report an error at corrections@runsonmy.com.
Confirmed errors
No reader-reported errors confirmed yet. Last automated check: Nov 17, 2026 (286 models checked, 0 held back). Pages held back by the checks are not published; see the methodology.
Changes to how the site works
- 2026-11-17: Fit pages for models that do not fit a card now say that llama.cpp CPU or expert offload (--n-cpu-moe, -ngl) can still run them more slowly, and that this is not modeled yet. Before, they only said "No".
- 2026-11-17: Search engines are now only offered model × GPU pages for popular models on consumer GPUs and Macs where the answer is close or depends on the settings. Every other pair still works; it is just marked noindex.
- 2026-11-17: Sitemap dates now show when a page's numbers last changed, not when the site was last rebuilt.
- 2026-11-17: Methodology wording: the golden tests use reference values computed independently from published model configs (we previously wrote "hand-computed"). Added a "What we don't model" section.
API price changes are logged separately on the price change log.