Model

Granite 4.0 H Small

OPEN

IBM · Granite 4.0 family · released Oct 2, 2025

Flagship hybrid Mamba-2/Transformer MoE (32B total, 9B active) cutting memory ~70%; first open models with ISO 42001 certification.

ReasoningCodingVisionFunction callingTool useAgentic

927.7

Elo · rank #228

Parameters

32B

Active params

9B (MoE)

Context

128K tokens

Architecture

Hybrid Mamba-2/Transformer MoE (4 attention / 36 Mamba2 layers; 72 experts, 10 active)

License

Apache 2.0

Languages

12+

API price (in/out)

$0.06 / $0.25

Modalities

text

Benchmark results

Bar shows position within the tracked field; marker = field best

AlpacaEvalHuman preference42.5%#25

best: Tülu 2+DPO 70B · 95.1%

Arena-HardHuman preference46.5%#43

best: Qwen3-235B-A22B (Non-Thinking) · 96.1%

BIG-Bench HardReasoning81.6%#29

best: ERNIE 4.5 300B-A47B · 94.3%

BFCL v3Agents64.7%#21

best: Hunyuan-A13B · 78.3%

BigCodeBenchCoding46.2%#4

best: GPT-4o mini · 57.4%

GPQA DiamondReasoning40.6%#215

best: GPT-5.6 · 94.6%

GSM8KMath87.3%#55

best: Llama 3.1 405B · 96.8%

Humanity's Last ExamReasoning4.0%#128

best: Claude Sonnet 5 · 57.4%

HumanEval+Coding83.0%#11

best: Mistral Small 3.2 (24B) · 92.9%

HumanEvalCoding88.0%#31

best: Claude Opus 4.5 · 99.4%

IFBenchReasoning31.0%#65

best: MiniMax M3 · 83.0%

IFEvalReasoning87.5%#39

best: Gemma 4 26B A4B · 98.5%

MBPP+Coding71.0%#15

best: Llama 3.1 405B · 88.6%

MBPPCoding84.0%#7

best: Llama-3.3-Nemotron-Super-49B v1 (Reasoning On) · 91.3%

MGSMMath38.7%#49

best: OpenAI o4-mini · 93.7%

MMLU-ProKnowledge55.5%#122

best: Claude Fable 5 · 91.5%

MMLUKnowledge78.4%#103

best: OpenAI o3 · 92.9%

MMMLUKnowledge69.7%#42

best: Gemini 3.1 Pro · 92.6%

Run it locally

VRAM @ Q4

21 GB

VRAM @ FP16

64.45 GB

Fits on (Q4)

RTX 3090 24GBRTX 4090 24GBRTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB

Throughput data unavailable.

Quantizations

GGUF Q4_K_M

Granite 4.0 family

Elo progression across releases

Granite 4.0 H SmallOct 2025927.7

Resources

HFibm-granite/granite-4.0-h-small GHibm-granite/granite-4.0-language-models DOCSibm.com/granite/docs/models/granite APIAvailable via IBM API

API price $0.06/$0.25 · each benchmark row carries its own source badge (see methodology)