Fetching from the wire…
01
02
03
04
05
06
07
08
Models2026-08-13 · source-backed
Cohere Labs released a native-resolution VLM that preserves aspect ratios instead of downsampling to a fixed grid, trained heavily on documents, charts, OCR, and grounding. Catch worth knowing before you wire it into a document pipeline: the language backbone carries 128K context but the validated multimodal range is only 8K tokens. Cohere positions it as a fine-tuning foundation, not a finished product.
Each link below shares sources, entities, or timing with this story.
Cohere released North Mini Code / Shared entities / Same source domain / Shared topic / Earlier coverage / Tension
Linked by a graph relationship (Cohere released North Mini Code); both cover Apache, Cohere, CohereLabs; reported by the same outlet (huggingface.co).
Shared entities / Same source domain / Shared topic / Earlier coverage
Both cover OCR, Vision; reported by the same outlet (huggingface.co); overlapping topics (chart, document).
Shared entities / Same source domain / Earlier coverage
Both cover Apache, Micro; reported by the same outlet (huggingface.co); earlier Apache coverage from 2026-07-26.
Both cover Apache, Micro; reported by the same outlet (huggingface.co); earlier Apache coverage from 2026-07-25.
Shared entities / Shared topic / Earlier coverage
Both cover Apache, Vision; overlapping topics (apache, context); earlier Apache coverage from 2026-03-22.
Shared entity: Apache / Same source domain / Shared topic / Earlier coverage
Both cover Apache; reported by the same outlet (huggingface.co); overlapping topics (apache, grounding).
Both cover Apache; reported by the same outlet (huggingface.co); overlapping topics (apache, context).
Both cover Apache; reported by the same outlet (huggingface.co); overlapping topics (apache, context).