Fetching from the wire…
01
02
03
04
05
06
07
08
Skills2026-03-29 · source-backed
At 85.5% exact-match accuracy (zero-shot) on OCR, chart analysis, and table parsing, this 3B model ranks 3rd among 2-4B VLMs. Trained on 32 H100 GPUs for ~200 hours. For document processing pipelines that don't need a frontier model, this is the best accuracy-per-compute option available.
Each link below shares sources, entities, or timing with this story.
Shared entities / Same source domain / Shared topic / What happened next
Both cover OCR, Vision; reported by the same outlet (huggingface.co); overlapping topics (chart, document).
Shared entity: OCR / Shared topic / What happened next
Both cover OCR; overlapping topics (document, extraction, frontier, model); picks up the OCR thread on 2026-06-30.
Shared entity: Vision / Same source domain / Shared topic / Earlier coverage
Both cover Vision; reported by the same outlet (huggingface.co); overlapping topics (hardware, model).
Shared entities / What happened next
Both cover OCR, Vision; picks up the OCR thread on 2026-08-09.
Same source domain / Shared topic / Tension
Reported by the same outlet (huggingface.co); overlapping topics (frontier, h100, model); pushes against this story (competes).
Shared entity: OCR / Shared topic / What happened next
Both cover OCR; overlapping topics (hour, model); picks up the OCR thread on 2026-08-14.
Shared entity: VLMs / Shared topic / What happened next
Both cover VLMs; overlapping topics (best, model); picks up the VLMs thread on 2026-07-30.
Shared entity: Trained / Same source domain / What happened next
Both cover Trained; reported by the same outlet (huggingface.co); picks up the Trained thread on 2026-07-28.