Model

Qwen2-72B

OPEN

Alibaba · Qwen2 family · released Jun 7, 2024

Gen-2 flagship, 7T training tokens, 128K context; topped the HF Open LLM Leaderboard on release.

ReasoningCodingVisionFunction callingTool useAgentic

832.7

Elo · rank #248

Parameters

72B

Active params

72B (dense)

Context

128K tokens

Architecture

Dense decoder-only Transformer (GQA)

License

Tongyi Qianwen License Agreement

Languages

30+

API price (in/out)

No hosted API

Modalities

text

Benchmark results

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

ARC-ChallengeReasoning68.9%#52

best: Llama 3.1 405B · 96.9%

Arena-HardHuman preference48.1%#41

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

BIG-Bench HardReasoning82.4%#26

best: ERNIE 4.5 300B-A47B · 94.3%

Gaokao MathMath54.6%#5

best: Doubao-Seed-1.6 · 96.0%

GPQA DiamondReasoning37.9%#230

best: GPT-5.6 · 94.6%

GSM8KMath89.5%#45

best: Llama 3.1 405B · 96.8%

HellaSwagReasoning87.6%#17

best: Claude 3 Opus · 95.4%

HumanEvalCoding64.6%#92

best: Claude Opus 4.5 · 99.4%

IFEvalReasoning77.6%#98

best: Gemma 4 26B A4B · 98.5%

LiveCodeBenchCoding35.7%#138

best: DeepSeek-V4-Pro (Think Max) · 93.5%

LogicKorHuman preference8.61#12

best: GPT-4o · 9.33

MATH-500Math51.1%#128

best: GPT-5 · 99.4%

MBPPCoding76.9%#22

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

MGSMMath86.6%#22

best: OpenAI o4-mini · 93.7%

MMLU-ProKnowledge55.6%#121

best: Claude Fable 5 · 91.5%

MMLUKnowledge84.2%#65

best: OpenAI o3 · 92.9%

MT-BenchHuman preference9.1#7

best: Hunyuan-Large (A52B) · 9.4

TruthfulQAKnowledge54.8%#38

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

WinoGrandeReasoning85.1%#11

best: PaLM 2 · 90.9%

Run it locally

VRAM @ Q4

42 GB

VRAM @ FP16

145 GB

Fits on (Q4)

M3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB

72B dense weights exceed a single RTX 4090's 24GB VRAM even at Q4.

Quantizations

GGUF · AWQ · GPTQ

Qwen2 family

Elo progression across releases

Qwen2-57B-A14BJun 2024589.2 Qwen2-72BJun 2024832.7 Qwen2-7BJun 2024485.5

Resources

HFQwen/Qwen2-72B-Instruct GHQwenLM/Qwen2 DOCSqwenlm.github.io/blog/qwen2

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