Granite 4.2 3B

Name: Granite 4.2 3B
Author: IBM (Granite)

IBM (Granite) · Granite 4.2 · released 2026-08-07 · data checked 2026-11-01

Granite 4.2 3B is a 3B-parameter dense model from IBM (Granite) (United States), released 2026-08-07 under the Apache-2.0 license. It scores 54.8 on the Small LM Index, #9 of 21 tiny models. It needs about 3 GB of memory at 4-bit with an 8K-token context.

Highest-scoring tiny (4B and under) open model on the Artificial Analysis Intelligence Index v5.

Small LM Index

54.8

#9 of 21 scored tiny models

Likely 51.2–58.2 · from 4 of 6 core benchmarks

Parameters

Dense

Price per 1M tokens

—

No hosted API on OpenRouter; self-host

Context window

131K

tokens

Benchmarks

Each bar shows this model's score. The tick marks the best tiny model on that benchmark. Sources: 7 reported by IBM (Granite). How we label sources.

MMLU-Pro Lab-reported Knowledge67.8

Class best: Qwen3.5 4B 79.1

GPQA Diamond Lab-reported Knowledge54.8

Class best: Nanbeige4.2 3B 87.4

AIME Lab-reported Math · AIME 202578.3

Class best: Spark-X2.6 4B 91.2

LiveCodeBench Lab-reported Coding · LiveCodeBench v669.7

Class best: LFM3 1.7B 73.5

IFBench Lab-reported Instruction following · Prompt-level74.3

Class best: Spark-X2.6 4B 75.8

BFCL Lab-reported Agentic · BFCL v452.4

Class best: Spark-X2.6 4B 71.8

HMMT Lab-reported Math · HMMT Feb 202566.7

Class best: Nanbeige4.2 3B 82.8

Not reported: estimated

What the index model expects from this model's other scores. Not a measurement.

Humanity's Last Exam≈ 8

SWE-bench Verified≈ 38

Specifications

Lab: IBM (Granite) (US · United States)
Architecture: Dense transformer
Total parameters: 3B
Active parameters: All (dense)
Size class: Tiny
Context window: 131K tokens
Input modalities: text
Reasoning: Instruct (no thinking)
License: Apache-2.0
Openness: Open weights
Memory to run: ~3 GB at 4-bit, 8K context (more sizes)
Hugging Face: 33.8K downloads (30 days) · 127 likes

Weights on Hugging Face ↗

Compare with

Granite 4.2 3B vs MiniCPM5 2B (2.5B, index 57.5)
Granite 4.2 3B vs Ministral 3 3B (3B, index 49.4)
Granite 4.2 3B vs SmolLM3 3B (3.1B, index 30.4)
Granite 4.2 3B vs LFM2.5 2.6B (2.7B, index 43.2)
Granite 4.2 3B vs Qwen3.5 2B (2.3B, index 33.0)

Ranked in

#9 in Best tiny LLMs for phones and edge devices

Sources

Something wrong here? Report it.

Run it locally

Uses ibm-granite/granite-4.2-3b-GGUF (Q4_K_M). Swap the tag for another size: BF16, Q2_K, Q3_K_L, Q3_K_M, Q3_K_S, Q4_0, Q4_1, Q4_K_M….

Ollama

ollama run hf.co/ibm-granite/granite-4.2-3b-GGUF:Q4_K_M

llama.cpp

llama-server -hf ibm-granite/granite-4.2-3b-GGUF:Q4_K_M

Then open http://localhost:8080 for a chat window.

MLX (Mac)

pip install -U mlx-lm
mlx_lm.chat --model lmstudio-community/granite-4.2-3b-MLX-4bit

vLLM (GPU server)

vllm serve ibm-granite/granite-4.2-3b

Full-precision weights, for a server GPU.

Open in LM Studio

Memory needed: weights, context cache and runtime buffers.

Quantization	4K context	32K context	131K context
16-bit F16	6.9 GB	9.2 GB	17 GB
8-bit Q8_0	4.1 GB	6.4 GB	14 GB
4-bit Q4_K_M	2.7 GB	5.1 GB	13 GB

Check it against your GPU, Mac or phone →