Granite 4.2 3B

+ contribute

Granite 4.2 3B is an open-weights language model from IBM, released August 2026. Its published size is 3.66B parameters and a 131k-token context window. Licence: apache-2.0.

57.2±0.8#231 of 719 LLMsfrom 35 benchmark results across 3 sources

$ how it compares

The LLMs ranked either side of Granite 4.2 3B, out of 828 with a score, then more from IBM.

#	model	score	where	RAM	released
274	Gemini 2.0 Flash Thinking Experimental (01-21) Google	57±2	comparison only		2025-01
275	DeepSeek R1 0528 Qwen3 8B DeepSeek	57±1	api $0.068/M +local	~6.3GB	2025-05
276	Granite 4.2 3B IBM	57±1	api $0.052/M +local	~3.7GB	2026-08
277	Command A (03-2025) Cohere	57±1	api $4.38/M +local	~37GB	2025-03
278	GPT-4o (2024-11-20) OpenAI	57±1	api $4.38/M		2024-11
	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
299	Granite 4.1 30B IBM	56±1	local	~19GB	2026-04

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #176 of 300 · 13% of its score · 6 tests

Beats o1 (2024-12-17), o3-mini, DeepSeek R1 0528 here, though they rank higher overall. Trails SmolLM3 3B, Jamba 1.5 Mini, Jamba 1.5 Large here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
BFCL V4 (December 2025 archive) +1	52.2% ✎	▄·█·▄▄▄▄·▄·▄ beats 5/8	medium · avg 43%	▇▇ 3.8%	at its level · several runs
SWE-bench Verified	36.8% ✎	▄··▄▄█··▄·▄█ beats 7/8	very hard · avg 12%	▇ 2.3%	sharp · noisy
BrowseComp-ZH	21.1% ✎	▄▄·▄▄▄·█···█ beats 6/8	very hard · avg 13%	▇ 2.3%	typical
SWE-bench Pro (split unspecified)	12.3% ✎	█▂···▂···▄·▂ beats 6/8	very hard · avg 4.3%	▇ 1.8%	sharp · noisy
τ²-bench Telecom	40.9% ✎	▁·▅█▆▅▂▃▃▂▃▁ beats 34/89	hard · avg 32%	▇ 1.4%	sharp · at its level · noisy · small run
τ³-bench lab	45.8% ✎	▁▁▂▁·▁▁·▃▁▂█ beats 8/25	medium · avg 37%	▇ 1.3%	blunt · at its level · small run

coding▁▁▁▃▆██▇▆▄▃▁ #226 of 458 · 21% of its score · 6 tests

Beats Qwen3.5 4B, Qwen3-VL 32B Instruct, Solar Open 100B (102B-A12B) here, though they rank higher overall. Trails QwQ-32B-Preview, Ministral 3 3B Reasoning 2512, GPT-4 (0613) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
LiveCodeBench v6 (window unspecified) +1	58.9% ✎	▃····█·▃▅▃·▃ beats 7/8	medium · avg 39%	▇▇▇ 8.0%	at its level · several runs
LiveCodeBench Pro 2025Q2 (Easy)	54.6% ✎	▄···█▄··▄█▄▄ beats 6/8	hard · avg 33%	▇ 3.6%	at its level
LiveCodeBench Pro 2025Q2 (Medium)	5.3% ✎	█···▂··▄▂··▂ beats 6/8	very hard · avg 2.0%	▇ 3.5%	sharp · too hard for it · noisy
SWE-bench Verified	36.8% ✎	▄··▄▄█··▄·▄█ beats 7/8	very hard · avg 12%	▇ 2.3%	sharp · noisy
SWE-bench Pro (split unspecified)	12.3% ✎	█▂···▂···▄·▂ beats 6/8	very hard · avg 4.3%	▇ 1.8%	sharp · noisy
SciCode (level unspecified) lab	24.1% ✎	▁····▃▄▄▄█▆▂ beats 6/34	hard · avg 19%	▇ 1.6%	at its level · small run

chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #227 of 393 · 9.2% of its score · 3 tests

Beats Granite 4.2 8B, Mistral Small 4 (119B-A6.5B), Olmo 3.1 32B Think here, though they rank higher overall. Trails Flan-T5 XXL, LFM2.5 1.2B Instruct, Gemma 3 4B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
LMArena (text)	1289.5 ✎	▁▁▁▂▃▃▃▆▆▇█▃ beats 123/379	avg 1268	▇▇ 5.6%	blunt
Arena-Hard v2 aggregate lab	35.0% ✎	▁▁··▄▃▁▃█▁▁▇ beats 3/22	hard · avg 27%	▇ 3.4%	sharp · at its level · small run
IFEval (aggregation unspecified)	93.7% ✎	▄▄····▄··███ beats 8/8	easy · avg 77%	▇ 0.2%	noisy

math▁▂▃▅▆▅▄█▄▁▁▁ #152 of 398 · 31% of its score · 5 tests

Beats MiMo-V2.5 (310B-A15B), MiniMax-M2.7, MiniMax-M2.1 here, though they rank higher overall.

test	result	vs the others in its run	difficulty	counts	why
AIME 2025 +1	79.4% ✎	█▄▄▄▄····█·▄ beats 7/8	medium · avg 59%	▇▇▇▇▇ 12%	sharp · small run · several runs
HMMT February 2026	60.8% ✎	▅··▃▃·▃··▃·█ beats 6/8	hard · avg 30%	▇▇▇ 8.2%	sharp · at its level
AIME 2026	83.5% ✎	▄·▄█·▄▄···█▄ beats 7/8	medium · avg 48%	▇▇ 5.0%	sharp · at its level
HMMT February 2025 lab	66.7% ✎	▁▁▁▃▁▅▅▃██▁▄ beats 8/31	medium · avg 57%	▇ 3.1%	at its level · small run
MATH-500	97.0% ✎	██▄▄▄··▄···▄ beats 7/8	very easy · avg 89%	▇ 2.6%	maxed out

web search█·█▄██·▄▄·▄▄ #6 of 13 * · 2.3% of its score · 1 test

Beats Qwen3.5 9B here, though they rank higher overall. Trails Spark-X2.5 1.7B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
BrowseComp-ZH	21.1% ✎	▄▄·▄▄▄·█···█ beats 6/8	very hard · avg 13%	▇ 2.3%	typical

other tests13 not in a task · 30% of its score

test	result	vs the others in its run	difficulty	counts	why
MMLU-Pro +1	65.8% ✎	▂··▂█·▂▂···▂ beats 5/8	medium · avg 59%	▇▇▇ 6.3%	at its level · several runs
GPQA (subset unspecified) lab	54.8% ✎	▂▁▂▂▄▃▆▃▅█▄▃ beats 56/104	hard · avg 42%	▇▇ 5.0%	sharp · at its level · small run
OJBench	21.8% ✎	▃····▃▅··█▃▃ beats 6/8	very hard · avg 12%	▇ 2.8%	typical
MMLU-Redux	78.9% ✎	▂···█··▂▂··▂ beats 1/8	easy · avg 77%	▇ 2.7%	typical
LongBench v2	36.0% ✎	▃······▅▅▃·█ beats 5/8	very hard · avg 32%	▇ 2.3%	typical
SuperGPQA	39.9% ✎	▃·▃·▃▅█····▃ beats 6/8	hard · avg 32%	▇ 2.2%	at its level
RULER 128k lab	55.3% ✎	▃··▅▄▇▅█▇·▃▃ beats 7/32	medium · avg 58%	▇ 1.5%	at its level · small run
IFBench prompt-level lab	74.3% ✎	█▄▄··▄·▄···▄ beats 3/6	medium · avg 62%	▇ 1.5%	at its level · small run

5 more tests

test	result	vs the others in its run	difficulty	counts	why
ProfBench lab	32.1% ✎	█·········██ beats 0/2	hard · avg 24%	▇ 1.4%	at its level · small run
IFBench (aggregation unspecified)	73.0% ✎	▅····▃▃·█·▃▃ beats 8/8	medium · avg 43%	▇ 1.4%	at its level
RULER 64k lab	67.5% ✎	▁▁·▃·▂▃▅█▂▁▁ beats 2/25	easy · avg 71%	▇ 1.2%	small run
MMLU-ProX-Lite lab	27.8% ✎	█·········██ beats 0/2	hard · avg 33%	▇ 1.0%	at its level · small run
Multi-IF (aggregation unspecified)	75.9% ✎	▄▄···▄·▄·▄██ beats 7/8	medium · avg 60%	▇ 0.9%	blunt · at its level

$ where to run it

Through an API

api · DeepInfra · $0.052/M · $0.03 in / $0.12 out per 1M tokens · 131k context · deepinfra.com ↗ · price sourceUSD per 1M text tokens. DeepInfra standard synchronous API, model ibm-granite/granite-4.2-3b; uncached input, no batch. Tools, cache storage and other modalities excluded. Provider does not specify quantization. Bertrande, https://bertrande.com/en/ai/deepinfra/ibm-granite--granite-4.2-3b; observed 2026-09-22T10:15:02Z. CC BY 4.0; units normalized and model identifier mapped.

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 4.2 3B · needs ~2.2 GB · from huggingface.co/ibm-granite
chat in the terminal
llama cli -hf ibm-granite/granite-4.2-3b-GGUF:Q4_K_M
Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf ibm-granite/granite-4.2-3b-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 4.2 3B · needs ~2.2 GB · from huggingface.co/ibm-granite
open in LM Studio ↗
Or download it in the terminal
lms get https://huggingface.co/ibm-granite/granite-4.2-3b-GGUF@Q4_K_M -y
Then run lms ls to see its name, and lms chat followed by that name

Ollama

Install Ollama
download it (macOS / Windows) ↗
Linux
curl -fsSL https://ollama.com/install.sh | sh
Run Granite 4.2 3B · needs ~2.7 GB · from ollama.com
chat in the terminal
ollama run granite4.2:3b

MLX (Mac)

version

Install MLX (Mac) · Apple silicon Macs only, text only.
with uv
uv tool install mlx-lm
or
pip install mlx-lm
Run Granite 4.2 3B · needs ~2.1 GB, Mac only · from huggingface.co/ibm-granite
chat in the terminal
mlx_lm.chat --model ibm-granite/granite-4.2-3b-q4-mlx
Or start an OpenAI-compatible server at http://localhost:8080/v1
mlx_lm.server --model ibm-granite/granite-4.2-3b-q4-mlx

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

$ how the score was worked out · openbmb/MiniCPM5-2B model card 59%, ibm-granite/granite-4.2-30b m… 36%, LMArena leaderboard dataset 6%

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
openbmb/MiniCPM5-2B model card	20	▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 59%	Apache-2.0
ibm-granite/granite-4.2-30b model card · lab-reported	14	▇▇▇▇▇▇▇▇▇▇▇ 36%	Apache-2.0
LMArena leaderboard dataset	1	▇▇ 6%	CC BY 4.0
prior · pull toward the middle		▇ 0%
weighted average		= 57.2

$ improve this page

Update model information Add a place to run it Add a test result Propose a benchmark Correct a benchmark score Edit 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 2026-10-11 02:20 UTC · sources and licences · json