whatmodel.sh / North Micro Vision Instruct (2.4B)
North Micro Vision Instruct (2.4B)
↗ share
+ contribute
North Micro Vision Instruct (2.4B) is an open-weights language model from Cohere, released August 2026. Its published size is 2.4B parameters and a 128k-token context window. Licence: apache-2.0.
Catalogue identity notes
Cohere's 2.4B vision-language model, CohereLabs/North-Micro-Vision-Instruct @fa548d9 (created 2026-08-10, Apache-2.0). Its own column in the card's Benchmark Results table, whose comparator columns are already recorded under card2-coherelabs-north-micro-vision-instruct.
Model information sources
Release date ↗ · Weights licence ↗
Model card: 'Total parameters 2.4B' (2B language model + 400M vision encoder).
$ how it compares
The LLMs ranked either side of North Micro Vision Instruct (2.4B), out of 828 with a score, then more from Cohere.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 611 |  Llama 2 70B (base) | 41±1 | local | ||
| 612 |  Mixtral 8x7B v0.1 (base) | 41±1 | local | ||
| 613 | North Micro Vision Instruct (2.4B) | 41±2 | local | ||
| 614 |  CodeQwen1.5 7B (base) | 41±1 | local | ||
| 615 |  Granite 4.0 1B | 41±1 | local | ||
| more from Cohere | |||||
| 76 | North Small Translate 1.0 | 71±6 | api price unknown | ||
| 175 | Command A+ (05-2026) | 64±1 | api price unknown +local | ||
| 277 | Command A (03-2025) | 57±1 | api $4.38/M +local | ||
$ what it's good at
chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #255 of 393 * · 6.8% of its score · 1 test
Beats Mistral Small 4 (119B-A6.5B), Ministral 3 14B Reasoning 2512, Ministral 3 14B here, though they rank higher overall. Trails Flan-T5 XXL, Granite 4.0 H 1B, Granite 4.0 1B here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| IFEval (aggregation unspecified) lab | 74.9% ✎ | ▄···▄█▄····▄ beats 4/5 | 6.8% |
vision & documents▁▁▃▂▁▃▄▂▆▇█▄ #136 of 147 · 56% of its score · 9 tests
Beats LLaVA-OneVision Qwen2 72B (ov-sft), Gemma 3 4B, CogVLM2 Llama3 Chat 19B here, though they rank higher overall. Trails Qwen3.5 0.8B here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| CharXiv descriptive questions lab | 60.0% ✎ | █·▄·······▄█ beats 1/5 | 19% |
| MMStar lab | 51.8% ✎ | ▄▄▄█·······▄ beats 2/5 | 12% |
| MMBench-EN dev v1.1 lab | 68.7% ✎ | █·····▃··▃·▃ beats 0/5 | 7.6% |
| HallusionBench lab | 61.5% ✎ | ▄▄·▄·····█·▄ beats 2/5 | 6.1% |
| CountBench lab | 72.5% ✎ | ▄··▄··█··▄·▄ beats 2/5 | 3.4% |
| RealWorldQA lab | 62.2% ✎ | ▄···█··▄▄··▄ beats 3/5 | 3.2% |
| OCRBench lab | 79.2% ✎ | █··███·█···█ beats 4/5 | 2.4% |
| AI2D test lab | 77.5% ✎ | █···▄▄···▄·▄ beats 4/5 | 1.6% |
$ where to run it
Other ways to run it
- Run it as published · needs ~2.17 GB, Mac only · from huggingface.co/mlx-community
pip install -U "mlx-vlm @ git+https://github.com/Blaizzy/mlx-vlm.git" && mlx_vlm.generate --model mlx-community/North-Micro-Vision-Instruct-4bit --image /path/to/image.jpg --prompt "Describe this image." --max-tokens 512 --temperature 0.0
Weights and documentation: huggingface.co/CohereLabs ↗
$ how the score was worked out · CohereLabs/North-Micro-Vision… 98%
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 CohereLabs/North-Micro-Vision-Instruct model card (benchmark table) · lab-reported | 13 | 98% |
| prior · pull toward the middle | 2% | |
| weighted average | = 40.7 |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json