[whatmodel.sh](../../README.md) / Granite 4.0 350M # Granite 4.0 350M IBM · released 2025-10 · 0.352B params · apache-2.0 **score 28.1** ±1.0 · #675 of 732 · from 24 benchmark results across 3 sources Score within the [LMArena (text)](https://whatmodel.sh/?set=lmarena-text&score=set) comparison set only. ## What it's good at ● marks Granite 4.0 350M among every model on that task or test, worst to best. 
task            vs the others          rank  counts  tests
agent tasks     ▁▁●▆█▄▅▅▄▄▂▁   #300 of 313*    3.6%      2
coding          ●··········█        #3 of 4     39%      7
chat & writing  ▁▁▁●▆▄▆█▆▇▇▃   #366 of 402*    7.4%      2
math            ▁▁●▅█▆▄█▄▂▁▁    #414 of 431     26%      6
agent tasks · #300 of 313 · 3.6% of its score Beats [QwQ-32B-Preview](qwq-32b-preview.md), [DeepSeek-Coder-V2 Lite Instruct (16B-A2.4B)](deepseek-coder-v2-lite-instruct.md), [Olmo 3 7B Instruct](olmo-3-7b-instruct.md) here, though they rank higher overall. 
test                             result  vs the others in its run   difficulty        counts  why
BFCL V4 (December 2025 archive)   19.0%  ▁▁▁●██▅▅▆▂▂▂ beats 4/77    hard · avg 31%      1.8%  typical
BFCL v3 (maker)                   39.3%  ▁▁▁▁●▃▅█▆▅▄▁ beats 14/171  medium · avg 62%    1.7%  blunt · at its level
coding · #3 of 4 · 39% of its score Beats [Granite 4.0 H 350M](granite-4-0-h-350m.md) here, though they rank higher overall. 
test                  result  vs the others in its run   difficulty           counts  why
HumanEval (maker)      39.0%  ▁▁▁▁▂▂●▆█▅▂▁ beats 64/205  very easy · avg 89%     12%  sharp · at its level
HumanEval+ (maker)     37.0%  ▁·▁▁▂▃●▄█▅▄▃ beats 28/145  very easy · avg 86%     10%  sharp · at its level
MBPP (maker)           48.0%  ▁▁▁▄▅●▆▅▄▁▂▁ beats 31/86   easy · avg 81%         4.9%  at its level
MBPP+ (maker)          38.0%  ▁··▁▁●▄█▆▄▁▁ beats 7/120   easy · avg 72%         3.4%  at its level
MultiPL-E (maker)      16.0%  ●··▄·▅▅▇▄█▃▄ beats 1/31    medium · avg 51%       3.3%  typical
BigCodeBench (maker)   11.1%  ●····▃·█▃▅█▃ beats 0/12    medium · avg 40%       3.1%  typical
CRUXEval-O (maker)     23.8%  ●▂██▅▂▁·▁·▁▁ beats 1/51    medium · avg 48%       1.9%  blunt
chat & writing · #366 of 402 · 7.4% of its score Beats [Code Llama 70B Instruct](codellama-70b-instruct.md), [Gemma 4 E2B](gemma-4-e2b.md), [Code Llama 34B Instruct](codellama-34b-instruct.md) here, though they rank higher overall. Trails [Gemma 3 270M IT](gemma-3-270m.md), [SmolLM2 360M Instruct](smollm2-360m.md) here, though they rank lower overall. 
test                                result  vs the others in its run  difficulty           counts  why
IFEval instruction-level strict (…   61.6%  ●▄····▄▄█▄·█ beats 0/8    very easy · avg 88%    3.8%  at its level
IFEval prompt-level strict (maker)   49.2%  ▁▁··●··▂▇█▃▃ beats 2/26   easy · avg 81%         3.5%  at its level
math · #414 of 431 · 26% of its score Beats [Command R (03-2024)](command-r-03-2024.md), [Llama 2 13B Chat](llama-2-13b-chat.md), [Flan-T5 Large](flan-t5-large.md) here, though they rank higher overall. Trails [GPT-Neo 125M](gpt-neo-125m.md), [OPT-125M](opt-125m.md), [GPT-2 XL (1.5B)](gpt-2-xl.md) here, though they rank lower overall. 
test                                result  vs the others in its run   difficulty            counts  why
GSM8K (maker)                        30.7%  ▂▂▂▂▃▄●▇██▆▄ beats 61/206  very easy · avg 89%      12%  sharp · at its level
GSM-Symbolic (maker)                 26.8%  ●▃····▃·▅▅█▃ beats 0/10    easy · avg 78%          4.4%  at its level
DeepMind Mathematics (maker)          8.5%  ●····▅·█▃▃·▅ beats 1/10    medium · avg 50%        3.7%  typical
Minerva MATH (maker)                 13.0%  ▁·▃▁●▂▁█▄▁▃▃ beats 5/27    medium · avg 59%        3.0%  typical
MGSM, 5 languages (en, es, fr, ja…   14.7%  ●·▄▄██···█·▄ beats 0/10    medium · avg 46%        2.4%  typical
OTIS mock AIME 2024–2025              0.3%  ●▃▄▃▃▃▇▇█▃▂▅ beats 2/197   very hard · avg 7.2%    0.0%  sharp · too hard for it · noisy
other tests (8) 
test                               result  vs the others in its run   difficulty            counts  why
BIG-Bench Hard (maker)              33.1%  ▁▁▁▁●▄▅▆█▃▃▁ beats 19/155  easy · avg 75%          6.9%  sharp
MMLU (maker)                        35.0%  ▁▁▁▁▁●▂▅██▆▄ beats 20/214  easy · avg 74%          6.7%  typical
GPQA (subset unspecified) (maker)   24.1%  ●▁▂▂▄▃▆▃▅█▃▃ beats 1/106   hard · avg 37%          3.3%  sharp · too hard for it
MMLU-Pro (maker)                    12.1%  ●▁·▁▁▁▃▂▄▅█▁ beats 0/188   medium · avg 46%        2.8%  too hard for it
INCLUDE (maker)                     27.7%  ●····▂▂▄▄▄█▄ beats 1/36    medium · avg 52%        2.4%  too hard for it
MMMLU (maker)                       28.2%  ●▁▁▂▃▂▄▄▅██▃ beats 1/46    medium · avg 56%        2.1%  typical
Chess Puzzles (Epoch)                0.0%  ●▁▁▁▁▂▃▃▃▂▁▁ beats 10/137  very hard · avg 5.8%    0.0%  sharp · too hard for it · noisy
GPQA Diamond                        11.2%  ▁▁▁▁▂▄▅▅███▄ beats 0/322   hard · avg 39%             —  failed run: Far below guessing. At 350M…
## How the score was worked out Each result implies a score on its own, after adjusting for how hard that test is and who ran it. The model's 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 is added up by who ran the tests. [Method](../method.md) 
source                                    results  share of the score
ibm-granite/granite-4.0-h-1b model card…       21   98% ▇▇▇▇▇▇▇▇▇▇▇▇
Berkeley Function Calling Leaderboard r…        1    2% ▇
Epoch AI                                        2    0% ▇
prior: pull toward the middle                        0% ▇
weighted average                                   = 28.1
Licences: ibm-granite/granite-4.0-h-1b model card ([Apache-2.0](https://huggingface.co/ibm-granite/granite-4.0-h-1b/blob/d18cca4c121edb87d022116d281ce212c9136f57/README.md)) · Berkeley Function Calling Leaderboard result archive ([Apache-2.0](../../licenses/bfcl-result-archive-v4-Apache-2.0.txt)) · Epoch AI ([CC BY 4.0](https://creativecommons.org/licenses/by/4.0/))
Not counted (1) 
source    test          score  why
Epoch AI  GPQA Diamond  11.2%  failed run: Far below guessing. At 350M this is more plausibly a format failure (no parseable letter) than knowledge 'worse than random'.
## 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 4.0 350M (Q4_K_M, needs ~0.2 GB, from [huggingface.co/ibm-granite](https://huggingface.co/ibm-granite/granite-4.0-350m-GGUF/tree/b8208a86a58427e1739265318028eb5895b74bf2)) ```sh llama cli -hf ibm-granite/granite-4.0-350m-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-4.0-350m-GGUF:Q4_K_M ``` Other versions: Q8_0 (~0.4 GB) `llama cli -hf ibm-granite/granite-4.0-350m-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 4.0 350M (Q4_K_M, needs ~0.2 GB, from [huggingface.co/ibm-granite](https://huggingface.co/ibm-granite/granite-4.0-350m-GGUF/tree/b8208a86a58427e1739265318028eb5895b74bf2)) ```sh lms get https://huggingface.co/ibm-granite/granite-4.0-350m-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 (~0.4 GB) `lms get https://huggingface.co/ibm-granite/granite-4.0-350m-GGUF@Q8_0 -y` · 4bit (~0.2 GB, Mac only) `lms get https://huggingface.co/mlx-community/granite-4.0-350m-4bit -y` · 8bit (~0.4 GB, Mac only) `lms get https://huggingface.co/mlx-community/granite-4.0-350m-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 4.0 350M (BF16, needs ~1.34 GB, from [ollama.com](https://ollama.com/library/granite4/tags)) ```sh ollama run granite4:350m ``` ### 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 4.0 350M (4bit, needs ~0.2 GB, Mac only, from [huggingface.co/mlx-community](https://huggingface.co/mlx-community/granite-4.0-350m-4bit/tree/f118e62936e8a845680a4c614997e14d7441f26e)) ```sh mlx_lm.chat --model mlx-community/granite-4.0-350m-4bit ``` 3. Or start an OpenAI-compatible server at http://localhost:8080/v1 ```sh mlx_lm.server --model mlx-community/granite-4.0-350m-4bit ``` Other versions: 8bit (~0.4 GB, Mac only) `mlx_lm.chat --model mlx-community/granite-4.0-350m-8bit` Model repository and documentation: [huggingface.co/ibm-granite](https://huggingface.co/ibm-granite/granite-4.0-350m) [interactive version](https://whatmodel.sh/model/granite-4-0-350m) · [JSON](https://whatmodel.sh/api?format=json&all=1)