whatmodel.sh / Granite 4.0 350M

Granite 4.0 350M

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Granite 4.0 350M is an open-weights language model from IBM, released October 2025. Its published size is 0.352B parameters and a 32k-token context window. Licence: apache-2.0.

Catalogue identity notes

Comparison model from the Epoch AI internal-evaluation CSVs (Epoch release date 2025-10-28). Dense Granite 4.0 350M (Nano family, October 2025); distinct from granite-4.0-h-350m. 352,379,904 parameters. ollama granite4:350m = 350m-bf16 (same digest).

25.9±1.1#669 of 719 LLMsfrom 24 benchmark results across 3 sources

$ how it compares

The LLMs ranked either side of Granite 4.0 350M, out of 828 with a score, then more from IBM.

#modelscorewhereRAMreleased
770GPT4All-13B-Snoozy Nomic AI26±2local~56GB2023-04
771OPT-13B (13 billion parameters) Meta26±1local~29GB2022-05
772Granite 4.0 350M IBM26±1local~1.7GB2025-10
773GPT-3 curie (API base model) OpenAI26±1comparison only2020-06
774Alpaca 13B Stanford CRFM26±2comparison only2023-03
more from IBM
192Granite 4.2 30B IBM63±1api $0.28/M +local~19GB2026-08
239Granite 4.2 8B IBM60±1api $0.11/M +local~6.6GB2026-08
276Granite 4.2 3B IBM57±1api $0.052/M +local~3.7GB2026-08

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #287 of 300 * · 3.5% of its score · 2 tests

Beats Olmo 3 7B Instruct, QwQ-32B-Preview, DeepSeek-Coder-V2 Lite Instruct (16B-A2.4B) here, though they rank higher overall.

testresultvs the others in its rundifficultycountswhy
BFCL V4 (December 2025 archive)19.0% ✎▁▁▁▃██▅▅▆▂▂▂ beats 4/77hard · avg 32%▇ 1.8%typical
BFCL v3 lab39.3% ✎▁▁▁▁▃▃▅█▆▅▄▁ beats 14/170medium · avg 61%▇ 1.7%blunt · at its level
coding█··········█ #3 of 4 · 39% of its score · 7 tests

Beats Granite 4.0 H 350M here, though they rank higher overall.

testresultvs the others in its rundifficultycountswhy
HumanEval lab39.0% ✎▁▁▁▁▁▃▅▅█▃▂▁ beats 52/190very easy · avg 90%▇▇▇▇▇ 12%sharp · at its level
HumanEval+ lab37.0% ✎▁·▁▁▂▃▄▄█▅▄▃ beats 28/145very easy · avg 88%▇▇▇▇ 10%sharp · at its level
MBPP lab48.0% ✎▁▁▁▄▅█▄▅▃▁▂▁ beats 27/81easy · avg 82%▇▇ 5.0%at its level
MultiPL-E lab16.0% ✎▃··▄·▅▅▇▄█▃▄ beats 1/31medium · avg 54%▇ 3.5%typical
MBPP+ lab38.0% ✎▁··▁▁▃▄█▆▄▁▁ beats 7/120easy · avg 73%▇ 3.5%at its level
BigCodeBench lab11.1% ✎▅····▃·█▃▅█▃ beats 0/12medium · avg 42%▇ 3.2%typical
CRUXEval-O lab23.8% ✎▃▂██▅▂▁·▁·▁▁ beats 1/51medium · avg 51%▇ 2.0%blunt
chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #358 of 393 * · 7.6% of its score · 2 tests

Beats Gemma 4 E2B, Code Llama 70B Instruct, Code Llama 34B Instruct here, though they rank higher overall. Trails Gemma 3 270M IT, SmolLM2 360M Instruct here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
IFEval instruction-level strict lab61.6% ✎▄▄····▄▄█▄·█ beats 0/8very easy · avg 89%▇▇ 4.0%at its level
IFEval prompt-level strict lab49.2% ✎▁▁··▃··▂▇█▃▃ beats 2/26easy · avg 81%▇ 3.5%at its level
math▃▂▃▅▆▅▄█▄▁▁▁ #393 of 398 · 25% of its score · 6 tests

Beats Llama 2 13B Chat, DeepSeek-Coder 1.3B Base, StarCoderBase 7B here, though they rank higher overall. Trails Qwen2 0.5B Instruct here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
GSM8K lab30.7% ✎▂▂▂▅▅▆▅▇█▇▄▃ beats 22/146very easy · avg 89%▇▇▇▇ 10%sharp · at its level
GSM-Symbolic lab26.8% ✎▃▃····▃·▅▅█▃ beats 0/10easy · avg 80%▇▇ 4.9%at its level
DeepMind Mathematics lab8.5% ✎▅····▅·█▃▃·▅ beats 1/10medium · avg 54%▇▇ 4.0%typical
Minerva MATH lab13.0% ✎▄·▃▂·██▆·▆▄▄ beats 2/21medium · avg 62%▇ 2.9%typical
MGSM, 5 languages (en, es, fr, ja, zh) lab14.7% ✎█·▄▄██···█·▄ beats 0/10medium · avg 49%▇ 2.5%typical
OTIS mock AIME 2024–20250.3% ✎▃▃▄▃▃▃▇▇█▃▂▅ beats 2/193very hard · avg 8.5%▇ 0.0%sharp · too hard for it · noisy
other tests8 not in a task · 24% of its score
testresultvs the others in its rundifficultycountswhy
BIG-Bench Hard lab33.1% ✎▁▁▁▁▃▄▅▅█▃▄▁ beats 11/129easy · avg 78%▇▇▇ 7.4%sharp
MMLU lab35.0% ✎▁▁▁▁▃▂▃▅█▃▆▃ beats 9/183easy · avg 75%▇▇ 6.0%typical
GPQA (subset unspecified) lab24.1% ✎▃▁▂▂▄▃▆▃▅█▄▃ beats 1/104hard · avg 38%▇ 3.3%sharp · too hard for it
MMLU-Pro lab12.1% ✎▃▁·▁▁▁▃▂▄▅█▁ beats 0/188medium · avg 49%▇ 2.8%typical
INCLUDE lab27.7% ✎▃····▂▂▄▄▄█▄ beats 1/36medium · avg 54%▇ 2.5%too hard for it
MMMLU lab28.2% ✎▃▁▁▂▃▂▄▄▅██▃ beats 1/46medium · avg 58%▇ 2.2%typical
Chess Puzzles (Epoch)0.0% ✎█▁▁▁▁▂▁▄▂▂▁▁ beats 10/135very hard · avg 5.9%▇ 0.0%sharp · too hard for it · noisy
GPQA Diamond11.2% ✎▁▁▁▁▂▄▅▅███▄ beats 0/310hard · avg 40%—failed run: Far below guessing. At 350M this is more plausibly a format failure (no parseable letter) than knowledge 'worse than random'.

$ where to run it

llama.cpp · recommended

version

  1. 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 ↗

  2. Run Granite 4.0 350M · needs ~0.2 GB · from huggingface.co/ibm-granite
    chat in the terminal
    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)
    llama serve -hf ibm-granite/granite-4.0-350m-GGUF:Q4_K_M

LM Studio

version

  1. Install LM Studio · Open the app once before using the lms command.

    download it (macOS / Windows / Linux) ↗

  2. Run Granite 4.0 350M · needs ~0.2 GB · from huggingface.co/ibm-granite

    open in LM Studio ↗

  3. Or download it in the terminal
    lms get https://huggingface.co/ibm-granite/granite-4.0-350m-GGUF@Q4_K_M -y
  4. Then run lms ls to see its name, and lms chat followed by that name

Ollama

  1. Install Ollama

    download it (macOS / Windows) ↗

    Linux
    curl -fsSL https://ollama.com/install.sh | sh
  2. Run Granite 4.0 350M · needs ~1.34 GB · from ollama.com
    chat in the terminal
    ollama run granite4:350m

MLX (Mac)

version

  1. Install MLX (Mac) · Apple silicon Macs only, text only.
    with uv
    uv tool install mlx-lm
    or
    pip install mlx-lm
  2. Run Granite 4.0 350M · needs ~0.2 GB, Mac only · from huggingface.co/mlx-community
    chat in the terminal
    mlx_lm.chat --model mlx-community/granite-4.0-350m-4bit
  3. Or start an OpenAI-compatible server at http://localhost:8080/v1
    mlx_lm.server --model mlx-community/granite-4.0-350m-4bit

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

$ how the score was worked out · ibm-granite/granite-4.0-h-1b… 98%, Berkeley Function Calling Lea… 2%, Epoch AI — internal evaluatio… 0%

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

sourceresultsshare of the scorelicence
ibm-granite/granite-4.0-h-1b model card · lab-reported21▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 98%Apache-2.0
Berkeley Function Calling Leaderboard result archive1▇ 2%Apache-2.0
Epoch AI — internal evaluations2▇ 0%CC BY 4.0
prior · pull toward the middle▇ 1%
weighted average= 25.9

not counted (1)

Epoch AI — internal evaluationsGPQA Diamond [correct]11.2%failed run: Far below guessing. At 350M this is more plausibly a format failure (no parseable letter) than knowledge 'worse than random'.

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit 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 · sources and licences · json