Cohere Open-Sources North Micro Vision, Its Smallest VLM

On August 13, Cohere officially released and open-sourced its smallest vision-language model to date, North Micro Vision. Featuring 2.4B parameters, the model's weights are publicly available on Hugging Face to deliver lightweight vision-language processing capabilities.

Confirmed

Point: The model has a total of 2.4B parameters, including a 2B language model backbone, and is open-sourced under the Apache 2.0 license.
Point: Core capabilities cover native-resolution vision processing, multi-turn image-text dialogue, spatial reasoning, visual grounding, and multilingual image understanding, with a particular strength in complex document understanding tasks.
Point: The company claims the model performs excellently across a wide range of vision understanding benchmarks, outperforming models like Gemma 4 E2B and Ministral 3.
Point: For deployment and fine-tuning, Cohere has partnered with NVIDIA to offer a solution based on NeMo AutoModel for NVIDIA GPUs, while also supporting custom fine-tuning via Axolotl.

Why it matters

Point: As a compact vision-language model, North Micro Vision retains multimodal capabilities while lowering deployment barriers, providing developers with a new open-source option for handling image-text tasks and document understanding in resource-constrained environments.

2026-08-13 ~ 2026-08-13 · 7 related posts

Primary sources

Cohere Releases North Micro Vision, a 2.4B Open-Source Vision-Language Model — cohere · 2026-08-13 00:07 (UTC+08:00)
Cohere Partners with NVIDIA and Axolotl to Enable Fine-Tuning for North Micro Vision — cohere · 2026-08-13 05:03 (UTC+08:00)
Cohere Open-Sources North Micro Vision Model Under Apache 2.0 — cohere · 2026-08-13 00:07 (UTC+08:00)

[source] Cohere Releases North Micro Vision, a 2.4B Open-Source Vision-Language Model — cohere · 2026-08-13
[source] Cohere Open-Sources North Micro Vision Model Under Apache 2.0 — cohere · 2026-08-13
Cohere Releases North Micro: 2.4B Open-Weight Vision-Language Model — pmttyji · 2026-08-13
Cohere Open-Sources North Micro Vision Model on Hugging Face — MaziyarPanahi · 2026-08-13
[source] Cohere Partners with NVIDIA and Axolotl to Enable Fine-Tuning for North Micro Vision — cohere · 2026-08-13
Cohere Releases North Micro Vision: Smallest VLM for Document Understanding, Apache 2.0 — aidangomez · 2026-08-13
Cohere Releases North-Micro-Vision Multimodal Model — CohereLabs · 2026-08-13