whatmodel.sh / Granite 3.0 2B Instruct

Granite 3.0 2B Instruct

+ contribute

Granite 3.0 2B Instruct is an open-weights language model from IBM, released October 2024. Its published size is 2.63B parameters and a 4k-token context window. Licence: apache-2.0.

Catalogue identity notes

Arena entry 'granite-3.0-2b-instruct' (text). LMArena's leaderboard table (arena-leaderboard space @6f2eeeb) labels 'granite-3.0-2b-instruct' as 'Granite-3.0-2B-Instruct' (IBM, Apache 2.0), linking https://huggingface.co/ibm-granite/granite-3.0-2b-instruct. Maker repo ibm-granite/granite-3.0-2b-instruct @5ad66c1: licence tag apache-2.0. 2,634,201,088 parameters from the Hugging Face safetensors metadata (dense, so active = total). Released 2024-10: month the maker's Hugging Face repo was created (createdAt 2024-10-02).

38.6±0.9#551 of 719 LLMsfrom 22 benchmark results across 5 sources

$ how it compares

The LLMs ranked either side of Granite 3.0 2B Instruct, out of 828 with a score, then more from IBM.

#	model	score	where	RAM	released
641	Falcon-H1 3B Instruct TII	39±4	local	~3.4GB	2025-05
642	Phi-2 (2.7B) Microsoft	39±1	local	~3.2GB	2023-12
643	Granite 3.0 2B Instruct IBM	39±1	local	~3.2GB	2024-10
644	Llama 2 70B Chat Meta	39±1	local	~32GB	2023-07
645	Code Llama 13B Instruct Meta	39±1	local	~9.8GB	2023-08
	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

coding▁▁▁▃▆██▇▆▄▃▁ #367 of 458 · 16% of its score · 6 tests

Beats LFM2.5 8B-A1B, Qwen3-VL 2B Thinking, DeepSeek R1 Distill Qwen 7B here, though they rank higher overall. Trails Qwen2.5 1.5B Instruct, WizardLM 13B V1.2 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
HumanEval lab	50.6% ✎	▄·▄█·▄·▄···█ beats 5/7	easy · avg 81%	▇▇▇▇ 8.9%	sharp · at its level
MBPP lab	41.0% ✎	▄···▄·▄▄█··█ beats 5/7	medium · avg 64%	▇▇ 4.3%	at its level
CodeReviewQA: solution identification (hard), invariant accuracy	14.2% ✎	▃··▁▂▃▃▂█▃▄▂ beats 11/47	medium · avg 54%	▇ 1.3%	sharp · noisy
CodeReviewQA: automated code refinement, exact match rate	20.4% ✎	▁▁▁▁▁▁▂▃▃█▅▄ beats 12/47	hard · avg 33%	▇ 1.0%	typical
CodeReviewQA: change localisation (hard), invariant accuracy	1.0% ✎	█▃▁▁▂▂▂▃▅▅▅▇ beats 9/47	medium · avg 36%	▇ 0.9%	sharp · noisy
CodeReviewQA: change type recognition, invariant accuracy	69.0% ✎	▁▁▁······▁▃█ beats 18/47	easy · avg 67%	▇ 0.0%	blunt · at its level · noisy

chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #341 of 393 * · 9.4% of its score · 2 tests

Beats Gemma 4 E2B, Code Llama 70B Instruct, QwQ-32B-Preview here, though they rank higher overall. Trails Gemma 3 270M IT, Granite 4.0 H 350M, LFM2 350M here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
LMArena (text)	1156.6 ✎	▁▁▁▂▃▃▃▆▆▇█▃ beats 45/379	avg 1268	▇▇▇ 6.7%	blunt
IFEval (aggregation unspecified) lab	46.1% ✎	▄····█·▄··██ beats 3/7	easy · avg 68%	▇ 2.6%	at its level

math▁▂▃▅▆▅▄█▄▁▁▁ #338 of 398 * · 14% of its score · 2 tests

Beats GPT-3.5 Turbo (0125), DBRX Instruct, Command R Plus (04-2024) here, though they rank higher overall. Trails Qwen3 0.6B, Gemma 7B (base), LFM2 700M here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
MATH (subset unspecified) lab	23.7% ✎	▄···▄▄▄▄··█▄ beats 4/7	medium · avg 53%	▇▇▇ 7.1%	at its level
GSM8K lab	59.7% ✎	█▄····▄··█▄▄ beats 5/7	easy · avg 81%	▇▇▇ 6.6%	sharp · at its level

other tests12 not in a task · 60% of its score

test	result	vs the others in its run	difficulty	counts	why
BIG-Bench Hard lab	43.7% ✎	▃▃···▃··█··▅ beats 4/7	medium · avg 67%	▇▇▇▇▇ 14%	sharp
WinoGrande lab	71.9% ✎	█···▄▄▄▄···█ beats 5/7	easy · avg 84%	▇▇▇ 8.2%	at its level
GPQA (subset unspecified) lab	28.6% ✎	▄·▄··▄··▄█·█ beats 3/7	hard · avg 39%	▇▇▇ 7.5%	sharp · too hard for it
MMLU lab	56.0% ✎	▄····▄·▄▄█·█ beats 3/7	easy · avg 77%	▇▇▇ 6.9%	at its level
MMLU-Pro lab	27.9% ✎	▄·····█··█▄█ beats 4/7	medium · avg 47%	▇▇ 5.5%	typical
HellaSwag lab	76.8% ✎	▄·▄··█▄·▄·▄▄ beats 5/7	easy · avg 87%	▇▇ 4.4%	at its level
BoolQ lab	84.9% ✎	▄▄····▄▄·█·█ beats 5/7	easy · avg 91%	▇▇ 4.3%	at its level
ARC Challenge lab	54.4% ✎	▄····█··▄▄▄█ beats 4/7	easy · avg 78%	▇ 3.6%	sharp · at its level · noisy

4 more tests

test	result	vs the others in its run	difficulty	counts	why
OpenBookQA lab	43.2% ✎	▄▄··▄··▄·█▄▄ beats 5/7	medium · avg 60%	▇ 3.6%	typical
EuroEval Life in the UK (English knowledge MC), test split, 5-shot, accuracy	71.3% ✎	▁▁▁▁▁▁▁▂▄█▄▃ beats 40/146	easy · avg 83%	▇ 1.1%	at its level · noisy
Social IQA lab	66.4% ✎	▃·····▃··█▃▅ beats 6/7	medium · avg 68%	▇ 0.8%	blunt · at its level
Vectara hallucination leaderboard, 2025 dataset (HHEM-2.3, factual consistency rate)	91.2% ✎	▁▁▂▃▄▄█▆▃▂▂▁ beats 28/140	very easy · avg 94%	▇ 0.7%	blunt · noisy

$ where to run it

llama.cpp · recommended

version

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 3.0 2B Instruct · needs ~1.6 GB · from huggingface.co/lmstudio-community
chat in the terminal
llama cli -hf lmstudio-community/granite-3.0-2b-instruct-GGUF:Q4_K_M
Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf lmstudio-community/granite-3.0-2b-instruct-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 3.0 2B Instruct · needs ~1.6 GB · from huggingface.co/lmstudio-community
open in LM Studio ↗
Or download it in the terminal
lms get https://huggingface.co/lmstudio-community/granite-3.0-2b-instruct-GGUF@Q4_K_M -y
Then run lms ls to see its name, and lms chat followed by that name

Ollama

version

Install Ollama
download it (macOS / Windows) ↗
Linux
curl -fsSL https://ollama.com/install.sh | sh
Run Granite 3.0 2B Instruct · needs ~2.08 GB · from ollama.com
chat in the terminal
ollama run granite3-dense:2b-instruct-q4_K_M

MLX (Mac)

Install MLX (Mac) · Apple silicon Macs only, text only.
with uv
uv tool install mlx-lm
or
pip install mlx-lm
Run Granite 3.0 2B Instruct · needs ~1.8 GB, Mac only · from huggingface.co/ekryski
chat in the terminal
mlx_lm.chat --model ekryski/granite-3.0-2b-instruct-4bit
Or start an OpenAI-compatible server at http://localhost:8080/v1
mlx_lm.server --model ekryski/granite-3.0-2b-instruct-4bit

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

$ how the score was worked out · IBM Granite 3.0 Language Mode… 88%, LMArena leaderboard dataset 7%, CodeReviewQA: The Code Review… 3% and 2 more

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 3.0 Language Models technical report (GitHub) · lab-reported	15	▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 88%	Apache-2.0
LMArena leaderboard dataset	1	▇▇ 7%	CC BY 4.0
CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)	4	▇ 3%	CC-BY-4.0
EuroEval leaderboards (results collection), English generative leaderboard	1	▇ 1%	MIT
Vectara hallucination leaderboard (vectara/hallucination-leaderboard repository)	1	▇ 1%	Apache-2.0
prior · pull toward the middle		▇ 0%
weighted average		= 38.6