Model

Granite 3.0 2B Instruct

OPEN

IBM · Granite 3.0 family · released Oct 21, 2024

Smaller dense Granite 3.0 model for cost-efficient enterprise inference; Apache 2.0, same 12T-token corpus, 12 languages.

ReasoningCodingVisionFunction callingTool useAgentic

181.6

Elo · rank #368

Parameters

2.53B

Active params

2.53B (dense)

Context

4K tokens

Architecture

Dense decoder-only transformer (GQA, RoPE, SwiGLU, RMSNorm)

License

Apache 2.0

Languages

12+

API price (in/out)

No hosted API

Modalities

text

Benchmark results

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

AGIEvalReasoning29.8%#37

best: OLMo 3-Think 32B · 88.2%

ARC-ChallengeReasoning54.4%#111

best: Llama 3.1 405B · 96.9%

BIG-Bench HardReasoning43.7%#107

best: ERNIE 4.5 300B-A47B · 94.3%

GPQA DiamondReasoning28.6%#267

best: GPT-5.6 · 94.6%

GSM8KMath59.7%#116

best: Llama 3.1 405B · 96.8%

HellaSwagReasoning76.8%#101

best: Claude 3 Opus · 95.4%

HumanEvalCoding50.6%#122

best: Claude Opus 4.5 · 99.4%

IFEvalReasoning46.1%#135

best: Gemma 4 26B A4B · 98.5%

MATH-500Math23.7%#175

best: GPT-5 · 99.4%

MBPPCoding41.0%#83

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

MGSMMath37.1%#50

best: OpenAI o4-mini · 93.7%

MMLU-ProKnowledge27.9%#166

best: Claude Fable 5 · 91.5%

MMLUKnowledge56.0%#218

best: OpenAI o3 · 92.9%

MT-BenchHuman preference7.66#53

best: Hunyuan-Large (A52B) · 9.4

OpenBookQAReasoning43.2%#40

best: Claude 1 · 90.8%

Social IQaReasoning66.4%#13

best: Apple DCLM-Baseline 7B · 82.9%

TruthfulQAKnowledge53.4%#44

best: Phi-3.5-MoE (16x3.8B, 6.6B active) · 77.5%

WinoGrandeReasoning71.9%#89

best: PaLM 2 · 90.9%

Run it locally

VRAM @ Q4

1.6 GB

VRAM @ FP16

5.27 GB

Fits on (Q4)

RTX 3060 12GBRTX 4070 Ti 16GBRTX 3090 24GBRTX 4090 24GBRTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB

Throughput data unavailable.

Quantizations

GGUF

Granite 3.0 family

Elo progression across releases

Granite 3.0 2B InstructOct 2024181.6 Granite 3.0 8B InstructOct 2024415.3

Resources

HFibm-granite/granite-3.0-2b-instruct GHibm-granite/granite-3.0-language-models DOCSibm.com/granite/docs APIAvailable via IBM API

API price weights · each benchmark row carries its own source badge (see methodology)