[whatmodel.sh](../../README.md) / Granite 4.0 H Tiny (7B-A1B)
# Granite 4.0 H Tiny (7B-A1B)
IBM · released 2025-09 · 6.94B params (1B active) · apache-2.0
**score 48.7** ±0.9 · #396 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 H Tiny (7B-A1B) among every model on that task or test, worst to best.
task vs the others rank counts tests
agent tasks ▄·●········█ #3 of 4* 1.4% 1
coding █···●······▄ #2 of 4 32% 7
chat & writing ▁▁▁▃▆▄▆●▆▇▇▃ #240 of 402* 4.7% 2
math ▁▁▃▅█●▄█▄▂▁▁ #247 of 431 25% 5
agent tasks · #3 of 4 · 1.4% of its score
Trails [Granite 4.0 Micro 3B](granite-4-0-micro.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
BFCL v3 (maker) 57.6% ▁▁▁▁▁▃▅●▆▅▄▁ beats 86/171 medium · avg 60% 1.4% blunt · at its level
coding · #2 of 4 · 32% of its score
test result vs the others in its run difficulty counts why
HumanEval (maker) 83.0% ▁▁▁▁▂▂▆▆█●▂▁ beats 185/205 easy · avg 84% 9.2% sharp
HumanEval+ (maker) 76.0% ▁·▁▁▂▃▄▄█●▄▃ beats 98/145 easy · avg 79% 8.3% sharp
MBPP (maker) 80.0% ▁▁▁▄▅█▆▅▄●▂▁ beats 80/86 easy · avg 78% 3.9% typical
BigCodeBench (maker) 41.1% ▅····▃·█▃▅●▃ beats 11/12 medium · avg 41% 3.1% at its level
MultiPL-E (maker) 55.8% ▃··▄·▅●▇▄█▃▄ beats 10/31 medium · avg 56% 3.1% at its level
MBPP+ (maker) 69.0% ▁··▁▁▃▄█●▄▁▁ beats 80/120 easy · avg 68% 2.7% at its level
CRUXEval-O (maker) 39.6% ▂▂█●▅▂▁·▁·▁▁ beats 23/51 medium · avg 46% 1.7% blunt · at its level
chat & writing · #240 of 402 · 4.7% of its score
Beats [Mistral Small 4 (119B-A6.5B)](mistral-small-4.md), [Qwen2.5 Coder 32B Instruct](qwen2-5-coder-32b.md), [Ministral 3 14B Reasoning 2512](ministral-3-14b-reasoning.md) here, though they rank higher overall. Trails [Flan-T5 XXL](flan-t5-xxl.md), [LFM2.5 1.2B Instruct](lfm2-5-1-2b-instruct.md), [Gemma 3 4B](gemma-3-4b.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
IFEval instruction-level strict (… 84.8% ▄▄····▄▄●▄·█ beats 4/8 very easy · avg 88% 2.4% typical
IFEval prompt-level strict (maker) 78.1% ▁▁··▂··▂●█▃▃ beats 7/26 easy · avg 80% 2.3% typical
math · #247 of 431 · 25% of its score
Beats [GPT-4o (2024-08-06)](gpt-4o-2024-08-06.md), [Command A (03-2025)](command-a.md), [GPT-4o (2024-05-13)](gpt-4o-2024-05-13.md) here, though they rank higher overall. Trails [Gemma 7B (base)](gemma-7b-base.md), [CodeGemma 7B (v1.0)](codegemma-7b.md), [Phi-2 (2.7B)](phi-2.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
GSM8K (maker) 84.7% ▂▂▂▂▃▄▅▇█●▆▄ beats 150/206 very easy · avg 86% 8.5% sharp
Minerva MATH (maker) 69.6% ▁·▃▁▁▂▁█▄▁●▃ beats 24/27 easy · avg 68% 5.4% at its level
DeepMind Mathematics (maker) 49.9% ▅····▅·●▃▃·▅ beats 6/10 medium · avg 53% 5.1% at its level
GSM-Symbolic (maker) 81.1% ▃▃····▃·▅▅●▃ beats 7/10 easy · avg 82% 4.1% typical
MGSM, 5 languages (en, es, fr, ja… 45.4% █·▄▄█●···█·▄ beats 7/10 medium · avg 44% 1.8% at its level
other tests (9)
test result vs the others in its run difficulty counts why
GPQA (subset unspecified) (maker) 32.6% ▂▁▂▂▄●▆▃▅█▃▃ beats 29/106 very hard · avg 35% 7.2% sharp
BIG-Bench Hard (maker) 66.3% ▁▁▁▁▂▄▅▆●▃▃▁ beats 103/155 easy · avg 74% 7.0% sharp · at its level
MMLU (maker) 68.7% ▁▁▁▁▁▂▂▅█●▆▄ beats 130/214 medium · avg 73% 6.7% at its level
MMLU-Pro (maker) 44.9% ▁▁·▁▁▁●▂▄▅█▁ beats 40/188 medium · avg 50% 5.0% at its level
INCLUDE (maker) 53.1% ▂····▂▂●▄▄█▄ beats 9/36 medium · avg 57% 4.2% at its level
MMMLU (maker) 61.9% ▂▁▁▂▃▂▄●▅██▃ beats 15/46 medium · avg 62% 3.7% at its level
EuroEval HellaSwag (English), tes… 70.4% ▃·▁▁▂▁▂▅●▇▄▁ beats 51/100 medium · avg 67% 1.6% sharp · at its level · noisy
EuroEval Life in the UK (English … 81.2% ▁▁▁▁▁▁▁▂▄●▄▃ beats 71/146 easy · avg 82% 1.1% at its level · noisy
Vectara hallucination leaderboard… 92.7% ▁▁▂▃●▄█▆▃▂▂▁ beats 41/140 very easy · avg 94% 0.5% blunt · maxed out · noisy
source results share of the score
ibm-granite/granite-4.0-h-small model c… 21 96% ▇▇▇▇▇▇▇▇▇▇▇▇
EuroEval leaderboards 2 3% ▇
Vectara hallucination leaderboard 1 1% ▇
prior: pull toward the middle 0% ▇
weighted average = 48.7
Licences: ibm-granite/granite-4.0-h-small model card ([Apache-2.0](../../licenses/ibm-granite--granite-4.0-h-small.txt)) · EuroEval leaderboards ([MIT](https://github.com/EuroEval/leaderboards/blob/1a77f1d2f8ba1ba05ab73287fb090d76bf41b2e0/LICENSE)) · Vectara hallucination leaderboard ([Apache-2.0](https://www.apache.org/licenses/LICENSE-2.0))
## 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 H Tiny (7B-A1B) (Q4_K_M, needs ~4.23 GB, from [huggingface.co/ibm-granite](https://huggingface.co/ibm-granite/granite-4.0-h-tiny-GGUF/tree/08d5a8a9741dd5c1a95d2d39e25253226aa1464e))
```sh
llama cli -hf ibm-granite/granite-4.0-h-tiny-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-h-tiny-GGUF:Q4_K_M
```
### 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 H Tiny (7B-A1B) (Q4_K_M, needs ~4.23 GB, from [huggingface.co/ibm-granite](https://huggingface.co/ibm-granite/granite-4.0-h-tiny-GGUF/tree/08d5a8a9741dd5c1a95d2d39e25253226aa1464e))
```sh
lms get https://huggingface.co/ibm-granite/granite-4.0-h-tiny-GGUF@Q4_K_M -y
```
3. Then run lms ls to see its name, and lms chat followed by that name
Other versions: 4bit (~3.9 GB, Mac only) `lms get https://huggingface.co/lmstudio-community/granite-4.0-h-tiny-MLX-4bit -y` · 8bit (~7.4 GB, Mac only) `lms get https://huggingface.co/lmstudio-community/granite-4.0-h-tiny-MLX-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 H Tiny (7B-A1B) (Q4_K_M, needs ~4.2 GB, from [ollama.com](https://ollama.com/library/granite4/tags))
```sh
ollama run granite4:tiny-h
```
### 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 H Tiny (7B-A1B) (4bit, needs ~3.9 GB, Mac only, from [huggingface.co/lmstudio-community](https://huggingface.co/lmstudio-community/granite-4.0-h-tiny-MLX-4bit/tree/a0f1fb8fdb387c4bd90e23e5e3f7ba9f7438b921))
```sh
mlx_lm.chat --model lmstudio-community/granite-4.0-h-tiny-MLX-4bit
```
3. Or start an OpenAI-compatible server at http://localhost:8080/v1
```sh
mlx_lm.server --model lmstudio-community/granite-4.0-h-tiny-MLX-4bit
```
Other versions: 8bit (~7.4 GB, Mac only) `mlx_lm.chat --model lmstudio-community/granite-4.0-h-tiny-MLX-8bit`
Model repository and documentation: [huggingface.co/ibm-granite](https://huggingface.co/ibm-granite/granite-4.0-h-tiny)
## Published speed measurements
Reference results for the hardware, engine and settings below. These are not verified speeds of the downloads above or predictions for your machine.
hardware precision speed engine and settings
Intel CoffeeLake-S GT2 [UHD… q4_k_m 11.55 tok/s ollama (version not recorded) · l…
Each speed links to its source, which has the full settings.
[interactive version](https://whatmodel.sh/model/granite-4-0-h-tiny) · [JSON](https://whatmodel.sh/api?format=json&all=1)