Model

Gemma 2B

OPEN

Google · Gemma family · released Feb 21, 2024

Smallest first-gen open Gemma; runs on modest/consumer hardware; superseded by Gemma 2/3/4 generations.

ReasoningCodingVisionFunction callingTool useAgentic

-150.7

Elo · rank #418

Parameters

2.5B

Active params

2.5B (dense)

Context

8K tokens

Architecture

Dense decoder-only Transformer (MQA, RoPE, GeGLU)

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

AGIEvalReasoning24.2%#40

best: OLMo 3-Think 32B · 88.2%

ARC-ChallengeReasoning42.1%#143

best: Llama 3.1 405B · 96.9%

ARC-EasyReasoning73.2%#40

best: Phi-3-medium (14B) · 97.7%

BIG-Bench HardReasoning35.2%#118

best: ERNIE 4.5 300B-A47B · 94.3%

GPQA DiamondReasoning22.7%#282

best: GPT-5.6 · 94.6%

GSM8KMath17.7%#167

best: Llama 3.1 405B · 96.8%

HellaSwagReasoning71.4%#120

best: Claude 3 Opus · 95.4%

HumanEvalCoding22.0%#166

best: Claude Opus 4.5 · 99.4%

IFEvalReasoning20.4%#153

best: Gemma 4 26B A4B · 98.5%

MATH-500Math11.8%#189

best: GPT-5 · 99.4%

MBPPCoding29.2%#99

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

MMLU-ProKnowledge13.7%#177

best: Claude Fable 5 · 91.5%

MMLUKnowledge42.3%#253

best: OpenAI o3 · 92.9%

OpenBookQAReasoning47.8%#31

best: Claude 1 · 90.8%

PIQAReasoning77.3%#59

best: GPT-4o mini · 93.1%

Social IQaReasoning49.7%#28

best: Apple DCLM-Baseline 7B · 82.9%

TriviaQAKnowledge53.2%#41

best: Sarvam-1 (2B) · 90.6%

WinoGrandeReasoning65.4%#120

best: PaLM 2 · 90.9%

Run it locally

VRAM @ Q4

2 GB

VRAM @ FP16

5 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

No verified RTX 4090 llama.cpp Q4 throughput number found for the original (Gemma 1) 2B checkpoint.

Quantizations

GGUF Q4_K_M · GGUF Q8_0

Gemma family

Elo progression across releases

Gemma 2BFeb 2024-150.7 Gemma 7BFeb 2024186.8

Resources

HFgoogle/gemma-2b GHgoogle-deepmind/gemma DOCSai.google.dev/gemma/docs/releases

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