With introducing the new "multiple languages" mul pseudo language (T285156), our existing evaluation of what makes for a Good Item changes. Previously, having a lot of labels was an indicator for high quality, this will be less the case once mul has been deployed.
This is a placeholder task for thinking about how and when to update our machine learning models.
WMF is also in the process of migrating away from ORES and to Liftwing, and in this context the Wikidata models are being reconsidered/retrained/modified/touched/something. This is currently ongoing. See T328813: Develop a ML-based service to detect vandalism on Wikidata and T321224: Wikidata Item Quality Model (including the discussions for both).