Model

Gemma 2 27B

OPEN

Google · Gemma 2 family · released Jun 27, 2024

Flagship dense 27B Gemma 2 with GQA and 13T-token training; rivals models more than twice its size.

ReasoningCodingVisionFunction callingTool useAgentic

628.2

Elo · rank #279

Parameters

27.2B

Active params

27.2B (dense)

Context

8K tokens

Architecture

Dense decoder-only Transformer, GQA + interleaved local/global attention, logit soft-capping

License

Gemma Terms of Use

Languages

—

API price (in/out)

No hosted API

Modalities

text

Benchmark results

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

AGIEvalReasoning55.1%#9

best: OLMo 3-Think 32B · 88.2%

ARC-ChallengeReasoning71.4%#45

best: Llama 3.1 405B · 96.9%

ARC-EasyReasoning88.6%#8

best: Phi-3-medium (14B) · 97.7%

BIG-Bench HardReasoning74.9%#43

best: ERNIE 4.5 300B-A47B · 94.3%

DROPReasoning74.2#40

best: Hunyuan-T1 · 93.1

GSM8KMath74.0%#94

best: Llama 3.1 405B · 96.8%

HellaSwagReasoning86.4%#26

best: Claude 3 Opus · 95.4%

HumanEvalCoding51.8%#119

best: Claude Opus 4.5 · 99.4%

MATH-500Math42.3%#147

best: GPT-5 · 99.4%

MBPPCoding62.6%#52

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

MMLUKnowledge75.2%#124

best: OpenAI o3 · 92.9%

PIQAReasoning83.2%#17

best: GPT-4o mini · 93.1%

Social IQaReasoning53.7%#15

best: Apple DCLM-Baseline 7B · 82.9%

TriviaQAKnowledge83.7%#9

best: Sarvam-1 (2B) · 90.6%

WinoGrandeReasoning83.7%#14

best: PaLM 2 · 90.9%

Run it locally

VRAM @ Q4

16 GB

VRAM @ FP16

54 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 · GGUF Q8 · AWQ · MLX

Gemma 2 family

Elo progression across releases

Gemma 2 27BJun 2024628.2 Gemma 2 9BJun 2024479.2

Resources

HFgoogle/gemma-2-27b GHgoogle-deepmind/gemma DOCSai.google.dev/gemma/docs/core/model_card_2

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