North Micro Vision Instruct

Name: North Micro Vision Instruct
Author: Cohere

Multimodal

Cohere

Compact open-weight vision-language model (2.4B) with native-resolution image support for VQA, captioning, grounding, OCR, charts, and documents. Custom 400M SigLIP 2-based vision encoder + 2B North Micro LLM backbone (Command A+ style). Multilingual and multi-image. LM context 128K; multimodal training validated to 8K. Not a reasoning/tool-calling model. Intended for prototyping and fine-tuning.

Key Specifications

Parameters

2.4B

Context

Release Date

August 12, 2026

Average Score

61.1%

Repository Model Weights Results Blog

Timeline

Key dates in the model's history

Announcement

August 12, 2026

Last Update

September 12, 2026

Today

September 20, 2026

Technical Specifications

Parameters

2.4B

Training Tokens

Knowledge Cutoff

Family

Capabilities

MultimodalZeroEval

Benchmark Results

Model performance metrics across various tests and benchmarks

General Knowledge

Tests on general knowledge and understanding

MMLU

HF model card table (MMLU test via VLMEvalKit) • Self-reported

50.4%

Multimodal

Working with images and visual data

AI2D

HF model card table (AI2D_TEST via VLMEvalKit) • Self-reported

77.5%

ChartQA

HF model card table (ChartQA Test via VLMEvalKit) • Self-reported

80.8%

DocVQA

HF model card table (DocVQA VAL via VLMEvalKit) • Self-reported

92.1%

Other Tests

Specialized benchmarks

BLINK

HF model card table (BLINK via VLMEvalKit) • Self-reported

52.7%

CharXiv-D

HF model card table (CharXiv DQ via VLMEvalKit) • Self-reported

60.0%

CountBench

HF model card table (CountBench via VLMEvalKit) • Self-reported

72.5%

Hallusion Bench

HF model card table (HallusionBench via VLMEvalKit) • Self-reported

61.5%

IFEval

HF model card table (IFEval via VLMEvalKit) • Self-reported

74.9%

InfoVQA

HF model card table (InfoVQA VAL via VLMEvalKit) • Self-reported

65.2%

MMBench-V1.1

HF model card table (MMBenchDEV_EN_V11 via VLMEvalKit) • Self-reported

68.7%

MMLU-Pro

HF model card table (MMLU-Pro test via VLMEvalKit) • Self-reported

30.7%

MMMU (val)

HF model card table (MMMU DEV_VAL via VLMEvalKit) • Self-reported

32.9%

MMStar

HF model card table (MMStar via VLMEvalKit) • Self-reported

51.8%

Multi-IF

HF model card table (Multi-If via VLMEvalKit) • Self-reported

37.3%

OCRBench

HF model card table (OCRBench via VLMEvalKit) • Self-reported

79.2%

OCRBench-V2 (en)

HF model card table (OCRBench v2_en via VLMEvalKit) • Self-reported

36.7%

RealWorldQA

HF model card table (RealWorldQA via VLMEvalKit) • Self-reported

62.2%

RefCOCO-avg

HF model card table (RefCOCO avg P@1 over 8 splits via VLMEvalKit) • Self-reported

73.2%

License & Metadata

License

apache_2_0

Announcement Date

August 12, 2026

Last Updated

September 12, 2026

Compare North Micro Vision Instruct

All comparisons

vs Seed 2.1 Turbo vs DeepSeek-V4-Pro-0813 vs Inkling vs Inkling-Small vs DeepSeek-V4.1-Flash vs Gemini 3.7 Flash vs IBM Granite 4.2 30B vs Claude Fable 5.1

Similar Models

All Models

Gemma 3n E2B

Google

MM8.0B

Best score:0.5 (ARC)

Released:Jun 2025

Gemma 3 4B

Google

MM4.0B

Best score:0.7 (HumanEval)

Released:Mar 2025

Price:$0.02/1M tokens

Gemma 3n E2B Instructed LiteRT (Preview)

Google

MM1.9B

Best score:0.7 (HumanEval)

Released:May 2025

Gemma 3n E4B Instructed

Google

MM8.0B

Best score:0.8 (HumanEval)

Released:Jun 2025

Price:$20.00/1M tokens

Gemma 3n E2B Instructed

Google

MM8.0B

Best score:0.7 (HumanEval)

Released:Jun 2025

Qwen2.5-Omni-7B

Alibaba

MM7.0B

Best score:0.8 (HumanEval)

Released:Mar 2025

Qwen3.5-2B

Alibaba

MM2.0B

Best score:0.5 (GPQA)

Released:Mar 2026

Command A+

Cohere

MM218.0B

Best score:0.8 (TAU)

Released:May 2026

Recommendations are based on similarity of characteristics: developer organization, multimodality, parameter size, and benchmark performance. Choose a model to compare or go to the full catalog to browse all available AI models.