whatmodel.sh / Qwen2 72B Instruct

Qwen2 72B Instruct

+ contribute

Qwen2 72B Instruct is an open-weights language model from Alibaba, released June 2024. Its published size is 72.71B parameters and a 131k-token context window. Licence: tongyi-qianwen.

Catalogue identity notes

Comparison model from the Epoch AI internal-evaluation CSVs (Epoch release date 2024-06-07). Instruct checkpoint Qwen2-72B-Instruct; distinct from Qwen2.5 72B. 72,706,203,648 parameters.

50.5±0.5#333 of 719 LLMsfrom 29 benchmark results across 20 sources

$ how it compares

The LLMs ranked either side of Qwen2 72B Instruct, out of 828 with a score, then more from Alibaba.

#	model	score	where	RAM	released
400	Llama-3-Instruct-8B-SimPO Princeton NLP	51±1	local	~6.6GB	2024-05
401	Athene-70B Nexusflow	51±1	local	~29GB	2024-07
402	Qwen2 72B Instruct Alibaba	51±1	local	~33GB	2024-06
403	Bielik 11B v2.0 Instruct SpeakLeash	51±3	local	~8.5GB	2024-08
404	OpenThinker3 7B Open Thoughts	50±1	local	~6GB	2025-05
	more from Alibaba
18	Qwen3.8 Max (2.4T-A95B) Alibaba	78±1	api $3.00/M +local	~380GB	2026-08
34	Qwen3.8 Max (0902) Alibaba	76±1	api $3.00/M		2026-09
37	Qwen3.8-Flash-Next Alibaba	75±1	local	~59GB	2026-08

see all LLMs · find the ones that fit your machine

$ what it's good at

coding▁▁▁▃▆██▇▆▄▃▁ #255 of 458 · 17% of its score · 4 tests

Beats Qwen3-VL 32B Instruct, Solar Open 100B (102B-A12B), GLM-4.7-Flash here, though they rank higher overall. Trails DeepSeek LLM 67B Chat, QwQ-32B-Preview, InternLM2.5 20B Chat here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
CRUXEval (ZeroEval, 800 problems)	59.1% ✎	▁▂▆▃▆█▆▃▂·▁▁ beats 19/47	medium · avg 61%	▇▇▇ 6.7%	sharp · at its level
LiveCodeBench official (Jan 1–Nov 1, 2024)	25.3% ✎	▁▁·▁▂▄▆█▅▂·▂ beats 11/31	hard · avg 26%	▇▇ 5.4%	sharp · at its level
Aider code editing (133 Exercism Python exercises, two attempts)	55.6% ✎	▁·▁▂▄▁▆▄█▆▂▂ beats 24/62	medium · avg 54%	▇▇ 4.6%	at its level
SciCode main problems (official leaderboard)	1.5% ✎	▃···█··▂·▃·▂ beats 4/17	very hard · avg 1.1%	▇ 0.7%	sharp · too hard for it

chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #263 of 393 · 22% of its score · 4 tests

Beats Ministral 3 14B Reasoning 2512, Ministral 3 14B, Phi-4 14B here, though they rank higher overall. Trails Flan-T5 XXL, Granite 4.0 H 1B, Granite 4.0 1B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
AlpacaEval 2 length-controlled	38.1% ✎	▁▂▂▃▆▆█▆▄▄▁▁ beats 94/122	medium · avg 39%	▇▇▇▇ 10%	at its level
Arena-Hard (v0.1, GPT-4-1106 judge)	46.9% ✎	▂▁▁▅█▅▄▃▃▅▃▁ beats 37/60	medium · avg 52%	▇▇▇ 6.7%	sharp · at its level
Polyrating rating (bias-aware re-fit of Chatbot Arena votes, mid-2024)	1182.0 ✎	▂▁▂▁▃▇█▆▇▄▂▂ beats 78/95	avg 1201	▇ 2.4%	blunt
LMArena (text)	1261.8 ✎	▁▁▁▂▃▃▃▆▆▇█▃ beats 99/379	avg 1268	▇ 2.4%	blunt

math▁▂▃▅▆▅▄█▄▁▁▁ #275 of 398 · 18% of its score · 6 tests

Beats Molmo 2 8B, DeepSeek-Coder-V2 Instruct (0724), Llama 3.1 70B Instruct here, though they rank higher overall. Trails DeepSeek R1 Distill Qwen 1.5B, Granite 8B Code Base (4K), Mixtral 8x7B v0.1 (base) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
MATH level 5	39.1% ✎	▂▂▅▆▄█▆▃▅▂▃▄ beats 30/92	hard · avg 32%	▇▇ 5.5%	sharp · at its level
MATH Level 5 (ZeroEval, 721 problems)	38.3% ✎	▁▁▁▆█▂▇▇▄··▂ beats 19/38	hard · avg 34%	▇ 3.4%	sharp · at its level
GSM8K (ZeroEval, zero-shot)	92.7% ✎	▁▄·▁▄█▅▅▅███ beats 27/43	very easy · avg 89%	▇ 2.7%	sharp
Minerva MATH lab	31.6% ✎	▃·····▃··█▅▅ beats 4/8	medium · avg 37%	▇ 2.7%	at its level
GSM8K lab	93.2% ✎	▄···█▄▄·▄·▄█ beats 5/8	very easy · avg 92%	▇ 1.9%	sharp · maxed out
MATH (subset unspecified) lab	69.0% ✎	▃····▃·▅·▅·█ beats 2/8	easy · avg 76%	▇ 1.4%	typical

other tests15 not in a task · 43% of its score

test	result	vs the others in its run	difficulty	counts	why
MMLU-Redux (ZeroEval, 2,778 questions)	81.6% ✎	▁·▁▁▄▄▄▅█▄▂▁ beats 33/53	easy · avg 77%	▇▇▇▇ 11%	at its level
LiveBench 2024-07-26	40.2% ✎	▁▁▁▂▂▃▂█▆▃▅▃ beats 45/61	medium · avg 41%	▇▇▇▇ 10%	at its level
ZebraLogic (ZeroEval, 1,000 puzzles)	21.4% ✎	▁▁▃▆██▁▁·▁▁▁ beats 33/57	hard · avg 18%	▇▇ 3.8%	sharp
BigCodeBench (Instruct) lab	38.2% ✎	▃▂·▆▆██▆▅▃▃▃ beats 21/33	medium · avg 41%	▇ 3.1%	at its level
GPQA Diamond	40.8% ✎	▁▁▁▁▂▄▆▄▅▇█▅ beats 41/208	hard · avg 40%	▇ 3.0%	sharp
BigCodeBench (Complete) lab	53.9% ✎	▂▁▁▄▂▅▇█▆▂▃▂ beats 54/58	medium · avg 50%	▇ 3.0%	at its level
WildBench v2 WB-Score (task macro)	44.4977 WB-Score ✎	▁·▁▁·▃█▄▇▆▅▆ beats 33/52	avg 45.199 WB-Score	▇ 2.7%	sharp
CPTUB average (SpeakLeash, 0-5 judge scale)	73.1% ✎	▁▂▂▂▃█▃▆▅▄▂▃ beats 63/95	easy · avg 66%	▇ 1.6%	at its level

7 more tests

test	result	vs the others in its run	difficulty	counts	why
Polish EQ-Bench v2 (SpeakLeash)	72.1% ✎	▁··▁▁▂▂▆▅▆█▆ beats 59/70	medium · avg 64%	▇ 1.3%	at its level
Open Persian LLM Leaderboard v2 average (21 tasks)	65.1% ✎	▂▄▂▃█▄▃▆▃▅▂▂ beats 64/81	medium · avg 58%	▇ 1.0%	blunt · at its level
KINA (Knowledge Index of Noah's Ark) accuracy, 899 ten-option items	24.9% ✎	▁▁▂▁▃▄█▆▄▅▂▃ beats 10/39	very hard · avg 22%	▇ 1.0%	blunt
Open PT-LLM Leaderboard average (9 Portuguese tasks)	79.1% ✎	▃▁▁▁▁▁▁▁▂█▅▁ beats 250/276	medium · avg 73%	▇ 0.5%	at its level · noisy
RULER 64k	79.8% ✎	▁▁·▁·▂▃▄█▂·▁ beats 7/22	easy · avg 83%	▇ 0.5%	typical
RULER 128k	53.7% ✎	▄··▂▄▃▆██▄▄▂ beats 5/21	easy · avg 68%	▇ 0.2%	at its level · noisy
Vectara hallucination leaderboard, 2025 dataset (HHEM-2.3, factual consistency rate)	95.3% ✎	▁▁▂▃▄▄█▆▃▂▂▁ beats 68/140	very easy · avg 94%	▇ 0.1%	blunt · maxed out · noisy

$ where to run it

llama.cpp · recommended

version

Install llama.cpp · Homebrew and winget installs name the commands llama-cli and llama-server instead.
macOS / Linux
curl -LsSf https://llama.app/install.sh | sh
Windows
irm https://llama.app/install.ps1 | iex
other ways to install ↗
Run Qwen2 72B Instruct · needs ~47.4 GB · from huggingface.co/Qwen (official)
chat in the terminal
llama cli -hf Qwen/Qwen2-72B-Instruct-GGUF:q4_k_m
Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf Qwen/Qwen2-72B-Instruct-GGUF:q4_k_m

LM Studio

version

Install LM Studio · Open the app once before using the lms command.
download it (macOS / Windows / Linux) ↗
Run Qwen2 72B Instruct · needs ~47.4 GB · from huggingface.co/Qwen (official)
open in LM Studio ↗
Or download it in the terminal
lms get https://huggingface.co/Qwen/Qwen2-72B-Instruct-GGUF@q4_k_m -y
Then run lms ls to see its name, and lms chat followed by that name

Ollama

version

Install Ollama
download it (macOS / Windows) ↗
Linux
curl -fsSL https://ollama.com/install.sh | sh
Run Qwen2 72B Instruct · needs ~44.13 GB · from ollama.com
chat in the terminal
ollama run qwen2:72b-instruct

MLX (Mac)

version

Install MLX (Mac) · Apple silicon Macs only, text only.
with uv
uv tool install mlx-lm
or
pip install mlx-lm
Run Qwen2 72B Instruct · needs ~40.9 GB, Mac only · from huggingface.co/mlx-community
chat in the terminal
mlx_lm.chat --model mlx-community/Qwen2-72B-Instruct-4bit
Or start an OpenAI-compatible server at http://localhost:8080/v1
mlx_lm.server --model mlx-community/Qwen2-72B-Instruct-4bit

Official weights and documentation: huggingface.co/Qwen ↗

$ how the score was worked out · WildEval ZeroEval published r… 27%, LiveBench model judgments (HF… 10%, AlpacaEval 1.0 and 2.0 leader… 10% and 17 more

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

source	results	share of the score	licence
WildEval ZeroEval published results	5	▇▇▇▇▇▇▇▇ 27%	Apache-2.0
LiveBench model judgments (HF livebench/model_judgment, 2024-10-22 revision) with the LiveBench question datasets	1	▇▇▇ 10%	Apache-2.0 (stated in the LiveBench datasheet)
AlpacaEval 1.0 and 2.0 leaderboards (tatsu-lab/alpaca_eval repository)	1	▇▇▇ 10%	Apache-2.0
Epoch AI — internal evaluations	2	▇▇▇ 9%	CC BY 4.0
Arena-Hard-Auto v0.1 leaderboard (repository README, pre-v2)	1	▇▇ 7%	Apache-2.0
BigCodeBench paper (Zhuo et al., ICLR 2025), arXiv 2406.15877v4 · lab-reported	2	▇▇ 6%	CC BY 4.0
Qwen (Alibaba) vendor-reported benchmark scores for its own models (qwen.ai launch posts, Qwen2.5-VL and Qwen3-VL technical reports, Qwen3.8-Flash-Next card) · lab-reported	3	▇▇ 6%	Factual vendor-reported score (not licensed content)
LiveCodeBench official website results (2024-12-26)	1	▇▇ 5%	CC-BY-SA-4.0
Aider code editing benchmark runs (old edit leaderboard)	1	▇ 5%	Apache-2.0
WildBench v2 leaderboard data (Ai2)	1	▇ 3%	Apache-2.0
Polyrating: A Cost-Effective and Bias-Aware Rating System for LLM Evaluation (arXiv:2409.00696v3)	1	▇ 2%	CC-BY-4.0
LMArena leaderboard dataset	1	▇ 2%	CC BY 4.0
SpeakLeash CPTUB (Complex Polish Text Understanding Benchmark)	1	▇ 2%	MIT
SpeakLeash Polish EQ-Bench leaderboard	1	▇ 1%	Apache-2.0
Open Persian LLM Leaderboard v2 (Part DP AI and AUT NLP Lab)	1	▇ 1%	Apache-2.0
Knowledge Index of Noah's Ark (arXiv:2606.05104v2)	1	▇ 1%	CC BY 4.0
SciCode leaderboard (repository README)	1	▇ 1%	Apache-2.0
NVIDIA RULER repository README results table	2	▇ 1%	Apache-2.0
Open Portuguese LLM Leaderboard (archived)	1	▇ 1%	Apache-2.0
Vectara hallucination leaderboard (vectara/hallucination-leaderboard repository)	1	▇ 0%	Apache-2.0
prior · pull toward the middle		▇ 0%
weighted average		= 50.5