Model

Granite 3.2 8B Instruct

OPEN

IBM · Granite 3.2 family · released Feb 26, 2025

Adds a toggleable 'thinking' chain-of-thought mode plus inference-scaling techniques that let it rival much larger models on math reasoning.

ReasoningCodingVisionFunction callingTool useAgentic

705.8

Elo · rank #265

Parameters

8.1B

Active params

8.1B (dense)

Context

128K tokens

Architecture

Dense decoder-only transformer with toggleable chain-of-thought (thinking=True/False)

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

AlpacaEvalHuman preference61.2%#12

best: Tülu 2+DPO 70B · 95.1%

Arena-HardHuman preference55.3%#40

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

BIG-Bench HardReasoning64.8%#73

best: ERNIE 4.5 300B-A47B · 94.3%

DROPReasoning50.95#54

best: Hunyuan-T1 · 93.1

GSM8KMath81.7%#73

best: Llama 3.1 405B · 96.8%

HumanEval+Coding85.7%#10

best: Mistral Small 3.2 (24B) · 92.9%

HumanEvalCoding89.3%#21

best: Claude Opus 4.5 · 99.4%

IFEvalReasoning74.3%#107

best: Gemma 4 26B A4B · 98.5%

MMLUKnowledge66.8%#169

best: OpenAI o3 · 92.9%

TruthfulQAKnowledge66.9%#8

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

Run it locally

VRAM @ Q4

6 GB

VRAM @ FP16

16.2 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 Q4_K_M · AWQ · GPTQ

Granite 3.2 family

Elo progression across releases

Granite 3.2 8B InstructFeb 2025705.8

Resources

HFibm-granite/granite-3.2-8b-instruct DOCSibm.com/granite/docs/models/granite APIAvailable via IBM API

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