LLaVA-OneVision: Easy Visual Task Transfer

Single-image, multi-image and video VLMs were separately built, so capabilities did not transfer between them.

Consolidates data, model and visual-representation lessons from the LLaVA-NeXT blog series into one architecture where a single model handles single-image, multi-image and video, with anyres-style visual encoding producing tokens that a single LM consumes.

Technical report, no peer-reviewed venue found. Relevant to inference because anyres multi-crop encoding is the dominant source of vision-token count in the LLaVA family: each additional tile adds a full patch grid to the LM prefill. It is also one of the most commonly deployed open VLMs in self-hosted serving stacks, so its token behaviour is what most people actually experience. VENUE_TRACK BACKFILL 2026-10-04 (paper-fields pass, agent w3p): venue_track 'unknown'. `venue` is 'arxiv-preprint', which means no peer-reviewed venue has been established for this work as of the verification date, so the track genuinely is not known. 'unknown' is used here rather than leaving the field null or guessing 'main': the schema separates 'unknown' from 'main' precisely so that an unprereviewed preprint is not silently counted as main-track work. This annotation is what lets a downstream filter distinguish 'peer reviewed, main track' from 'not peer reviewed' without inferring it from the venue string.