whatmodel.sh / Granite 4.0 H Micro 3B

Granite 4.0 H Micro 3B

+ contribute

Granite 4.0 H Micro 3B is an open-weights language model from IBM, released September 2025. Its published size is 3.19B parameters and a 131k-token context window. Licence: apache-2.0.

Catalogue identity notes

IBM Granite-4.0-H-Micro (ibm-granite/granite-4.0-h-micro @d5f01a3, created 2025-09-16, apache-2.0, 3,191,396,096 params): hybrid Mamba-2/attention dense 3B instruct model; distinct from the catalogued transformer-only granite-4-0-micro (ollama granite4:micro). Lab rows from the h-small card table's 'H Micro Dense' column; independent EuroEval and Vectara rows.

Model information sources

Release date ↗ · Weights licence ↗

HF safetensors metadata of ibm-granite/granite-4.0-h-micro @d5f01a3

46.5±0.9#410 of 719 LLMsfrom 24 benchmark results across 3 sources

$ how it compares

The LLMs ranked either side of Granite 4.0 H Micro 3B, out of 828 with a score, then more from IBM.

#	model	score	where	RAM	released
486	InternLM2.5 7B Chat 1M Shanghai AI Laboratory (InternLM)	47±4	local	~6.4GB	2024-06
487	EuroLLM 9B Instruct UTTER (EuroLLM)	47±3	local	~6.9GB	2024-11
488	Granite 4.0 H Micro 3B IBM	47±1	local	~3.4GB	2025-09
489	GPT-3.5 Turbo (0125) OpenAI	46±1	api $0.75/M		2024-01
490	Mistral Large (2402) Mistral	46±1	comparison only		2024-02
	more from IBM
192	Granite 4.2 30B IBM	63±1	api $0.28/M +local	~19GB	2026-08
239	Granite 4.2 8B IBM	60±1	api $0.11/M +local	~6.6GB	2026-08
276	Granite 4.2 3B IBM	57±1	api $0.052/M +local	~3.7GB	2026-08

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▄·▄········█ #4 of 4 * · 1.3% of its score · 1 test

Trails Granite 4.0 Micro 3B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
BFCL v3 lab	57.6% ✎	▁▁▁▁▁▃▅█▆▅▄▁ beats 84/170	medium · avg 60%	▇ 1.3%	blunt · at its level

coding█···▄······▄ #4 of 4 · 34% of its score · 7 tests

Trails Granite 4.0 Micro 3B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
HumanEval lab	81.0% ✎	▁▁▁▁▁▃▅▅█▃▂▁ beats 162/190	very easy · avg 86%	▇▇▇▇ 10%	sharp
HumanEval+ lab	75.0% ✎	▁·▁▁▂▃▄▄█▅▄▃ beats 96/145	easy · avg 81%	▇▇▇▇ 8.9%	sharp · at its level
MBPP lab	73.0% ✎	▁▁▁▄▅█▄▅▃▁▂▁ beats 70/81	easy · avg 79%	▇▇ 4.1%	at its level
BigCodeBench lab	37.9% ✎	▅····▃·█▃▅█▃ beats 8/12	medium · avg 42%	▇ 3.1%	at its level
MultiPL-E lab	49.5% ✎	▃··▄·▅▅▇▄█▃▄ beats 6/31	medium · avg 58%	▇ 3.1%	at its level
MBPP+ lab	64.0% ✎	▁··▁▁▃▄█▆▄▁▁ beats 59/120	easy · avg 70%	▇ 2.8%	at its level
CRUXEval-O lab	41.3% ✎	▂▂██▅▂▁·▁·▁▁ beats 30/51	medium · avg 48%	▇ 1.7%	blunt · at its level

chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #206 of 393 * · 4.9% of its score · 2 tests

Beats Granite 4.2 8B, Qwen3-VL 30B-A3B Thinking, Mistral Small 4 (119B-A6.5B) here, though they rank higher overall. Trails LFM2.5 1.2B Instruct here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
IFEval instruction-level strict lab	86.9% ✎	▄▄····▄▄█▄·█ beats 6/8	very easy · avg 88%	▇ 2.6%	typical
IFEval prompt-level strict lab	81.7% ✎	▁▁··▂··▂▇█▃▃ beats 10/26	easy · avg 81%	▇ 2.4%	typical

math▁▂▃▅▆▅▄█▄▁▁▁ #266 of 398 · 24% of its score · 5 tests

Beats Molmo 2 8B, Llama 3.3 70B Instruct, DeepSeek-V2.5 here, though they rank higher overall. Trails DeepSeek R1 Distill Qwen 1.5B, Mixtral 8x7B v0.1 (base), InternVL2-4B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
GSM8K lab	81.3% ✎	▂▂▂▅▅▆▅▇█▇▄▃ beats 94/146	very easy · avg 87%	▇▇▇ 7.3%	sharp
DeepMind Mathematics lab	43.8% ✎	▅····▅·█▃▃·▅ beats 4/10	medium · avg 55%	▇▇ 5.4%	at its level
Minerva MATH lab	66.4% ✎	▄·▂▂·██▆·▆▄▄ beats 16/21	easy · avg 70%	▇▇ 5.2%	at its level
GSM-Symbolic lab	77.5% ✎	▃▃····▃·▅▅█▃ beats 5/10	easy · avg 83%	▇▇ 4.5%	typical
MGSM, 5 languages (en, es, fr, ja, zh) lab	44.5% ✎	█·▄▄██···█·▄ beats 6/10	medium · avg 46%	▇ 1.8%	at its level

other tests9 not in a task · 35% of its score

test	result	vs the others in its run	difficulty	counts	why
BIG-Bench Hard lab	69.4% ✎	▁▁▁▁▂▄▅▅█▃▄▁ beats 88/129	easy · avg 76%	▇▇▇ 7.5%	sharp · at its level
GPQA (subset unspecified) lab	32.1% ✎	▂▁▂▂▄▃▆▃▅█▄▃ beats 27/104	hard · avg 37%	▇▇▇ 6.5%	sharp
MMLU lab	67.4% ✎	▁▁▁▁▁▂▃▅█▃▆▃ beats 98/183	easy · avg 74%	▇▇ 6.0%	at its level
MMLU-Pro lab	43.5% ✎	▁▁·▁▁▁▃▂▄▅█▁ beats 34/188	medium · avg 52%	▇▇ 4.7%	at its level
INCLUDE lab	50.5% ✎	▂····▂▃▄▄▄█▄ beats 5/36	medium · avg 58%	▇▇ 4.3%	at its level
MMMLU lab	55.2% ✎	▂▁▁▂▃▂▄▄▅██▃ beats 12/46	medium · avg 63%	▇ 3.6%	at its level
EuroEval HellaSwag (English), test split, 5-shot generative MC, accuracy	67.7% ✎	▃·▁▁▂▁▂▅█▇▄▁ beats 43/100	medium · avg 68%	▇ 1.3%	at its level · noisy
EuroEval Life in the UK (English knowledge MC), test split, 5-shot, accuracy	81.6% ✎	▁▁▁▁▁▁▁▂▄█▄▃ beats 73/146	easy · avg 83%	▇ 1.0%	at its level · noisy

1 more test

test	result	vs the others in its run	difficulty	counts	why
Vectara hallucination leaderboard, 2025 dataset (HHEM-2.3, factual consistency rate)	91.6% ✎	▁▁▂▃▄▄█▆▃▂▂▁ beats 32/140	very easy · avg 94%	▇ 0.5%	blunt · maxed out · noisy

$ where to run it

llama.cpp · recommended

Install llama.cpp · Homebrew and winget installs name the commands llama-cli and llama-server instead.
macOS / Linux
curl -LsSf https://llama.app/install.sh | sh
Windows
irm https://llama.app/install.ps1 | iex
other ways to install ↗
Run Granite 4.0 H Micro 3B · needs ~1.94 GB · from huggingface.co/ibm-granite
chat in the terminal
llama cli -hf ibm-granite/granite-4.0-h-micro-GGUF:Q4_K_M
Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf ibm-granite/granite-4.0-h-micro-GGUF:Q4_K_M

LM Studio

version

Install LM Studio · Open the app once before using the lms command.
download it (macOS / Windows / Linux) ↗
Run Granite 4.0 H Micro 3B · needs ~1.94 GB · from huggingface.co/ibm-granite
open in LM Studio ↗
Or download it in the terminal
lms get https://huggingface.co/ibm-granite/granite-4.0-h-micro-GGUF@Q4_K_M -y
Then run lms ls to see its name, and lms chat followed by that name

Ollama

Install Ollama
download it (macOS / Windows) ↗
Linux
curl -fsSL https://ollama.com/install.sh | sh
Run Granite 4.0 H Micro 3B · needs ~1.9 GB · from ollama.com
chat in the terminal
ollama run granite4:micro-h

MLX (Mac)

version

Install MLX (Mac) · Apple silicon Macs only, text only.
with uv
uv tool install mlx-lm
or
pip install mlx-lm
Run Granite 4.0 H Micro 3B · needs ~1.8 GB, Mac only · from huggingface.co/mlx-community
chat in the terminal
mlx_lm.chat --model mlx-community/granite-4.0-h-micro-4bit
Or start an OpenAI-compatible server at http://localhost:8080/v1
mlx_lm.server --model mlx-community/granite-4.0-h-micro-4bit

Weights and documentation: huggingface.co/ibm-granite ↗

$ how the score was worked out · ibm-granite/granite-4.0-h-sma… 97%, EuroEval leaderboards (result… 2%, Vectara hallucination leaderb… 1%

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	licence
ibm-granite/granite-4.0-h-small model card · lab-reported	21	▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 97%	Apache-2.0
EuroEval leaderboards (results collection), English generative leaderboard	2	▇ 2%	MIT
Vectara hallucination leaderboard (vectara/hallucination-leaderboard repository)	1	▇ 1%	Apache-2.0
prior · pull toward the middle		▇ 0%
weighted average		= 46.5

$ improve this page

Update model information Add a place to run it Add a test result Propose a benchmark Correct a benchmark score Edit data on GitHub

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 2026-10-11 02:20 UTC · sources and licences · json