[whatmodel.sh](../../README.md) / Granite 3.3 2B Instruct
# Granite 3.3 2B Instruct
IBM · released 2025-04 · 2.53B params · apache-2.0
**score 41.0** ±1.3 · #549 of 732 · from 9 benchmark results across 1 source · † no independently run results verified
Score within the [LMArena (text)](https://whatmodel.sh/?set=lmarena-text&score=set) comparison set only.
## What it's good at
● marks Granite 3.3 2B Instruct among every model on that task or test, worst to best.
task vs the others rank counts tests
chat & writing ▁▁▁▃●▄▆█▆▇▇▃ #321 of 402* 13% 2
math ▁▁▃▅●▆▄█▄▂▁▁ #305 of 431 45% 3
chat & writing · #321 of 402 · 13% of its score
Beats [Gemma 4 E4B](gemma-4-e4b.md), [OpenThinker3 7B](openthinker3-7b.md), [Code Llama 70B Instruct](codellama-70b-instruct.md) here, though they rank higher overall. Trails [LFM2 350M](lfm2-350m.md), [Gemma 3 1B](gemma-3-1b.md), [Qwen3 0.6B](qwen3-0-6b.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
IFEval (aggregation unspecified) … 65.8% ▄▄·●······▄█ beats 2/5 easy · avg 78% 6.6% at its level
Arena-Hard (version/judge unspeci… 28.9% █·●··▄·····█ beats 2/5 medium · avg 55% 6.2% at its level
math · #305 of 431 · 45% of its score
Beats [Molmo 2 8B](molmo-2-8b.md), [Llama 3.1 Nemotron 70B Instruct](llama-3-1-nemotron-70b.md), [GPT-4 (0613)](gpt-4-0613.md) here, though they rank higher overall. Trails [DeepSeek R1 Distill Qwen 1.5B](deepseek-r1-distill-qwen-1-5b.md), [Gemma 7B (base)](gemma-7b-base.md), [Mistral 7B v0.1 (base)](mistral-7b-v0-1.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
GSM8K (maker) 72.5% █··●·····▄·█ beats 2/5 very easy · avg 89% 28% sharp · at its level
AIME 2024 (maker) 3.3% █···▄▄●····▄ beats 4/5 very hard · avg 8.9% 8.9% sharp · too hard for it
MATH-500 (maker) 58.1% █···▄·▄●···▄ beats 4/5 easy · avg 70% 7.8% at its level
other tests (4)
test result vs the others in its run difficulty counts why
BIG-Bench Hard (maker) 63.9% █·●·····▄▄·▄ beats 2/5 easy · avg 81% 19% sharp · at its level
MMLU (maker) 55.9% ●█········▄█ beats 0/5 easy · avg 76% 16% at its level
PopQA (maker) 18.4% ●·█······▄·█ beats 0/5 hard · avg 33% 6.0% typical
DROP (maker) 44.3% ▃▃·····●···█ beats 2/5 medium · avg 54% 0.8% blunt · at its level
source results share of the score
ibm-granite/granite-3.3-8b-instruct mod… 9 99% ▇▇▇▇▇▇▇▇▇▇▇▇
prior: pull toward the middle 1% ▇
weighted average = 41.0
Licences: ibm-granite/granite-3.3-8b-instruct model card ([Apache-2.0](../../licenses/ibm-granite--granite-3.3-8b-instruct.txt))
## How to run it
### llama.cpp (recommended)
1. Install llama.cpp. Homebrew and winget installs name the commands llama-cli and llama-server instead. [Other ways to install](https://github.com/ggml-org/llama.cpp/blob/master/docs/install.md)
```sh
# macOS / Linux
curl -LsSf https://llama.app/install.sh | sh
# Windows
irm https://llama.app/install.ps1 | iex
```
2. Run Granite 3.3 2B Instruct (Q4_K_M, needs ~1.5 GB, from [huggingface.co/ibm-granite](https://huggingface.co/ibm-granite/granite-3.3-2b-instruct-GGUF/tree/7cdf86ccd1f1bb3491c9b7017b033f2e51367397))
```sh
llama cli -hf ibm-granite/granite-3.3-2b-instruct-GGUF:Q4_K_M
```
3. Or start a local server and open the address it prints (http://localhost:9931)
```sh
llama serve -hf ibm-granite/granite-3.3-2b-instruct-GGUF:Q4_K_M
```
Other versions: Q8_0 (~2.7 GB) `llama cli -hf ibm-granite/granite-3.3-2b-instruct-GGUF:Q8_0`
### LM Studio
1. Install LM Studio. Open the app once before using the lms command. [Download it (macOS / Windows / Linux)](https://lmstudio.ai/download)
2. Run Granite 3.3 2B Instruct (Q4_K_M, needs ~1.5 GB, from [huggingface.co/ibm-granite](https://huggingface.co/ibm-granite/granite-3.3-2b-instruct-GGUF/tree/7cdf86ccd1f1bb3491c9b7017b033f2e51367397))
```sh
lms get https://huggingface.co/ibm-granite/granite-3.3-2b-instruct-GGUF@Q4_K_M -y
```
3. Then run lms ls to see its name, and lms chat followed by that name
Other versions: Q8_0 (~2.7 GB) `lms get https://huggingface.co/ibm-granite/granite-3.3-2b-instruct-GGUF@Q8_0 -y` · 4bit (~1.4 GB, Mac only) `lms get https://huggingface.co/mlx-community/granite-3.3-2b-instruct-4bit -y` · 8bit (~2.7 GB, Mac only) `lms get https://huggingface.co/mlx-community/granite-3.3-2b-instruct-8bit -y`
### Ollama
1. Install Ollama. [Download it (macOS / Windows)](https://ollama.com/download)
```sh
# Linux
curl -fsSL https://ollama.com/install.sh | sh
```
2. Run Granite 3.3 2B Instruct (Q4_K_M, needs ~2.02 GB, from [ollama.com](https://ollama.com/library/granite3.3/tags))
```sh
ollama run granite3.3:2b
```
### MLX (Mac)
1. Install MLX (Mac). Apple silicon Macs only, text only.
```sh
# with uv
uv tool install mlx-lm
# or
pip install mlx-lm
```
2. Run Granite 3.3 2B Instruct (4bit, needs ~1.4 GB, Mac only, from [huggingface.co/mlx-community](https://huggingface.co/mlx-community/granite-3.3-2b-instruct-4bit/tree/58246c5498495c14599525c852cfadb66c9f3084))
```sh
mlx_lm.chat --model mlx-community/granite-3.3-2b-instruct-4bit
```
3. Or start an OpenAI-compatible server at http://localhost:8080/v1
```sh
mlx_lm.server --model mlx-community/granite-3.3-2b-instruct-4bit
```
Other versions: 8bit (~2.7 GB, Mac only) `mlx_lm.chat --model mlx-community/granite-3.3-2b-instruct-8bit`
Model repository and documentation: [huggingface.co/ibm-granite](https://huggingface.co/ibm-granite/granite-3.3-2b-instruct)
[interactive version](https://whatmodel.sh/model/granite-3-3-2b) · [JSON](https://whatmodel.sh/api?format=json&all=1)