Qwen2-72B

72.7B parameters, text in, text out, 128K tokens of context. 1 provider serves it.

Verified 2026-10-25 Rank #615 in LLMs · score 73.8 method v1 Sources 3 Downloads 30d 24.1k · #441 of 619 tracked

Can I run it?44GB at INT4Estimatefits a 64 GB Mac (unified memory)method

Can I use it?tongyi-qianwenCustom license; read the repo LICENSE before commercial hostinglicence file2026-10-24

Do people use it?24.1kdownloads, 30 days#441 of 619 we track · below the median+2 likes, downloads −16% this week206 likesHugging Face2026-10-24

Worth reading

Picked by rule from 8 tracked pieces: the lab's own post, the most shared, a field report, the latest.

Most discussed · 29 HN pts

Alibaba Qwen2.5-Omni-7B: Open Source End-to-End Multimodal AI Model

alizila.com 2025-03-27 29 HN pts · 3 comments about the Qwen2 family family

Latest · 2 years ago

Qwen2-Math

Qwen2-Math is a series of specialized math language models built upon the Qwen2 LLMs that significantly outperforms the mathematical capabilities of open-source models and even closed-source models.

The model series primarily supports English, with bilingual English and Chinese math models scheduled for release soon.
Evaluation of multi-choice benchmarks including MMLU STEM and GaoKao problems uses a 5-shot setting instead of zero-shot.

qwenlm.github.io 2024-08-08 128 HN pts · 38 comments 5.8k words about the Qwen2 family family

8 best of 8 tracked

The 2 picks above, plus these 6. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.

ArticleQwen2 LLM Releasedqwenlm.github.io · 2024-06-06 · 261 HN pts · official
CodeLlama.cpp Now Supports Qwen2-VL (Vision Language Model)github.com · 2024-12-14
ArticleAn Analysis of Chinese LLM Censorship and Bias with Qwen 2 Instructhuggingface.co · 2024-06-09 · 16 HN pts
ArticleQwen2.5-Max: The Next Leap in AI Innovationmedium.com · 2025-01-29 · 17 HN pts
ArticleQwen2 72B, Gemini Flash, GPT-4, Opus Added to PVQ Leaderboardpvq.app · 2024-06-12 · 4 HN pts
ArticleQwen2-7B-Instruct with TensorRT-LLM: consistently high tokens/SECinferless.com · 2024-09-05

Show all 6

Where to run it

Live hosted endpoints, cheapest first, USD per million tokens. Ties break on output price, then provider name.

#	Cheapest live row	In /1M	Out /1M	Context	Quant	Seen
1	Featherless	—	—	—	—	Oct 25

All 1 provider →

Where it's used

Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.

Served by 1 API provider; 9 quantized builds for local use.

As a service1Featherless
On own hardware9quantized builds
Built on2423 fine-tunes · 1 adapters · 0 merges
In apps9Hugging Face Spaces

Top Spaces running it: Duongkum999/Qwen-Qwen2-72B ♥ 2 · Pyboxs/hf-llm-api ♥ 1

Will it fit?

Weights only, an estimate: leave room for the KV cache. Method

Fits a 64 GB Mac (unified memory) at INT4

About 44 GB at INT4, 87 GB at FP8, 175 GB at BF16: weights only, so leave room for the KV cache.

Estimate params × bytes per param × 1.2 · how we estimate

The story so far

Weights, every dated article, and repo commits on one rail.

6 dated articles and 12 repo commits so far

Teal is the weights (2024-05-22), amber triangles are 6 dated articles (each links to it), grey ticks are repo commits.

What people built on it

33 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.

33 derivatives on the Hub, most of them fine-tunes

From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.

GGUF bartowski/dolphin-2.9.2-qwen2-72b-GGUF for llama.cpp, Ollama, LM Studio 13.7k downloads MLX mlx-community/dolphin-vision-72b-4bit for Apple silicon 13.2k downloads FP8 novita-ai/dolphin-2.9.2-qwen2-72b-fp8 for vLLM, SGLang 4.8k downloads

All variants →

Reference

Everything else, folded: the full tables, every source.

Benchmarks

17 rows

Benchmarks 17 rows

measured third-party leaderboard rowauthor-reported number from the model card or eval files (17)No confidence intervals unless the source gave them; settings differ between rows, so compare within a benchmark, not across. Date = the benchmark's own date when it gives one, else when we read it.

Benchmark	Value	Setting / notes	Reported by	Date
author-reported ARC Challenge · Chat and follow instructions	68.9	—	model author	2026-10-07
author-reported BBH · Reason through hard problems	82.4	—	model author	2026-10-07
author-reported C-Eval · Work across languages	91	—	model author	2026-10-07
author-reported CMMLU · Work across languages	90.1	—	model author	2026-10-07
author-reported GPQA Diamond · Reason through hard problems	37.9	—	model author	2026-10-07
author-reported GSM8K · Reason through hard problems	89.5	—	model author	2026-10-07
author-reported HellaSwag · Chat and follow instructions	87.6	—	model author	2026-10-07
author-reported HumanEval · Write code with agents	64.6	—	model author	2026-10-07
author-reported HumanEval · Write code with agents	65.4	—	model author	2026-10-07
author-reported MATH-500 · Reason through hard problems	51.1	—	model author	2026-10-07
author-reported MBPP · Write code with agents	76.9	—	model author	2026-10-07
author-reported MMLU · Reason through hard problems	84.2	—	model author	2026-10-07
author-reported MMLU-Pro · Reason through hard problems	55.6	—	model author	2026-10-07
author-reported MultiPL-E · Write code with agents	59.6	—	model author	2026-10-07
author-reported Theorem QA	43.1	—	model author	2026-10-07
author-reported TruthfulQA · Chat and follow instructions	54.8	—	model author	2026-10-07
author-reported WindoGrande	85.1	—	model author	2026-10-07

What the card says

1 cited sentence

Summary generated, cited

The model uses a dense architecture based on the Transformer architecture with SwiGLU activation and group query attention.^[1]

“Transformer architecture with SwiGLU activation, attentio…”

Generated 2026-10-07 15:15 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.

Use-case evidence

5 use cases

What it is good at 5 use cases

Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.

Reason through hard problemsauthor-reportedAuthor reports MMLU 84.2 Author reports MMLU-Pro 55.6 Author reports GPQA 37.9 +3 more
Chat and follow instructionsauthor-reportedAuthor reports HellaSwag 87.6 Author reports ARC-C 68.9 Author reports TruthfulQA 54.8 +2 more
Write code with agentsauthor-reportedAuthor reports EvalPlus 65.4 Author reports MBPP 76.9 Author reports MultiPL-E 59.6
Work across languagesauthor-reportedAuthor reports C-Eval 91.0 Author reports CMMLU 90.1
Read long documentsstructuralconfig: 131,072 context

Facts

21 fields

Listed in LLMs

Facts

Qwen2-72B has 72.7B parameters. (safetensors metadata, observed 2026-10-24)
Context length is 131,072 tokens (128K). (config.json max_position_embeddings, observed 2026-10-24)
License: tongyi-qianwen (friction: review). Custom license; read the repo LICENSE before commercial hosting. (license file, observed 2026-10-24)
Created 2024-05-22, last modified 2024-10-09. (HF API, observed 2026-10-24)

Pipeline tag	text-generation
Library	transformers
Architectures	Qwen2ForCausalLM
Model type	qwen2
Dtype	bfloat16
Params by dtype	BF16: 72,706,203,648
Size band	XL
MoE	no
Context length	131,072 (config.json max_position_embeddings)
Languages	en
License	tongyi-qianwen (other)
Hub snapshot	2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: supported · Qwen2ForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · Qwen2ForCausalLM · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed

vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.

Hardware fit

3 estimates

Hardware fit ESTIMATE

Weights-only memory for Qwen2-72B: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)	175 GB · Multi-GPU node (<640 GB)
FP8 (1 byte/param)	87 GB · Multi-GPU node (<640 GB)
INT4 (0.5 bytes/param)	44 GB · Single high-end GPU (<80 GB)

Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.

All variants

33 variants

Lineage

Original model Family: Qwen/Qwen2 33 derivatives on the Hub

finetune: 23quantized: 9adapter: 1

Top variants (33 shown, ranked by downloads)

Variant	Relation	Quant	Params	Band	License	Author	Likes	Downloads
allenai/Molmo-72B-0924	finetune	—	73.3B	XL	apache-2.0	allenai	300	95.5k
dphn/dolphin-2.9.2-qwen2-72b	finetune	—	72.7B	XL	other	dphn	182	79.7k
dphn/dolphin-vision-7b	finetune	—	8.06B	M	other	dphn	57	37.6k
dphn/dolphin-vision-72b	finetune	—	73.2B	XL	other	dphn	136	11.1k
MaziyarPanahi/calme-2.2-qwen2-72b	finetune	—	72.7B	XL	other	MaziyarPanahi	5	17.1k
bartowski/dolphin-2.9.2-qwen2-72b-GGUF	quantized	GGUF	—	—	other	bartowski	7	13.7k
mlx-community/dolphin-vision-72b-4bit	finetune	-4bit	73.2B	XL	other	mlx-community	8	13.2k
MaziyarPanahi/calme-2.3-qwen2-72b	finetune	—	72.7B	XL	other	MaziyarPanahi	2	12.7k
mradermacher/Qwen2-72B-i1-GGUF	quantized	GGUF	—	—	other	mradermacher	0	9k
mradermacher/Qwen2-72B-GGUF	quantized	GGUF	—	—	other	mradermacher	1	6.7k
novita-ai/dolphin-2.9.2-qwen2-72b-fp8	quantized	-fp8	72.7B	XL	other	novita-ai	1	4.8k
AI-MO/NuminaMath-72B-CoT	finetune	—	72.7B	XL	other	AI-MO	7	1.5k
featherless-ai-quants/Qwen-Qwen2-72B-GGUF	quantized	GGUF	—	—	—	featherless-ai-quants	0	1.6k
mlx-community/dolphin-2.9.2-qwen2-72b-8bit	finetune	-8bit	20.5B	L	other	mlx-community	2	592
mlx-community/dolphin-2.9.2-qwen2-72b-4bit	finetune	-4bit	11.4B	L	other	mlx-community	1	419
BlouseJury/cognitivecomputations_dolphin-2.9.2-qwen2-72b-Q6_K-gguf	quantized	Q6_K	—	—	other	BlouseJury	0	400
mlx-community/dolphin-2.9.2-qwen2-72b-2bit	finetune	—	6.82B	M	other	mlx-community	0	393
atcsecure/dolphin-2.9.2-qwen72b-8.0bpw-h8-exl2	quantized	0bpw	—	—	other	atcsecure	2	245
BlouseJury/cognitivecomputations_dolphin-2.9.2-qwen2-72b-6.0bpw-exl2	finetune	0bpw	—	—	other	BlouseJury	0	327
migtissera/Tess-v2.5-Qwen2-72B-QLoRA-Adapter	adapter	—	—	—	—	migtissera	4	109
bigstorm/dolphin-2.9.2-qwen2-72b-6.0bpw-exl2	quantized	0bpw	—	—	other	bigstorm	1	256
bigstorm/dolphin-2.9.2-qwen2-72b-7.0bpw-8hb-exl2	quantized	0bpw	—	—	other	bigstorm	2	184
awseeley/test-dol	finetune	—	8.06B	M	other	awseeley	0	186
tf919/Qwen2-72B-Instruct-test	finetune	—	72.7B	XL	other	tf919	0	73
zeromodels/qwen2-72b	finetune	—	—	—	other	zeromodels	0	53
indigoskyai/Molmo-72B-0924	finetune	—	73.3B	XL	apache-2.0	indigoskyai	0	49
godofconquest/Molmo-72B-0924	finetune	—	73.3B	XL	apache-2.0	godofconquest	0	46
glockinhand/insomnia_0.0.1_Qwen2_72B	finetune	—	72.7B	XL	other	glockinhand	0	44
vennilaml/Molmo-72B-0924	finetune	—	73.3B	XL	apache-2.0	vennilaml	0	44
user888888888/dolphin-2.9.2-qwen2-72b	finetune	—	72.7B	XL	other	user888888888	0	39
Loskoso/dolphin-2.9.2-qwen2-72b	finetune	—	72.7B	XL	other	Loskoso	0	36
fiztag/dolphin-2.9.2-qwen2-72b	finetune	—	72.7B	XL	other	fiztag	0	34
DonkeyErika/dolphin-vision4-donkeys	finetune	—	73.2B	XL	other	DonkeyErika	0	6

Repository changelog

12 commits

Repository changelog 12 commits

Date	Commit	SHA
2024-10-09	Expose metadata link to next version of the model (#9)	5f149ca
2024-06-06	update README.md	8799379
2024-06-06	Update config.json	42848b7
2024-06-06	Update README.md	c03367e
2024-05-31	Update tokenizer_config.json	e5cebed
2024-05-31	Update README.md	e7f6075
2024-05-31	Update LICENSE	f7f6228
2024-05-30	update config.json & tokenizer_config.json	5d27038
2024-05-30	Update README.md	a60aebe
2024-05-30	Update README.md	14d7f0a
2024-05-22	Upload folder using huggingface_hub	999320f
2024-05-22	initial commit	83edd4a

All providers

1 provider · 1 endpoint

Where to run 1 providers · 1 endpoints

Hosted endpoints observed for Qwen2-72B, sorted by input price (USD per million tokens). Each row names its observer and date; prices change daily, so check the source before quoting.

Provider	Model ref	Context	Quant	$/M in	$/M out	tok/s	TTFT	Tools	Status	Observed
Featherless hf	Qwen/Qwen2-72B	—	—	—	—	—	—	—	live	2026-10-25

Articles and coverage

8 pieces

Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.

Official3

Qwen2 LLM Released

qwenlm.github.io2024-06-06officialauthorityHN front pageabout the Qwen2 familyHN 261 pts · 130 comments4,713 words

Alibaba releases Qwen2, a new series of open-source models featuring five sizes ranging from 0.5B to 72B parameters. The update applies Group Query Attention to all sizes and trains on data in 27 additional languages besides English and Chinese. Specific instruction-tuned versions like Qwen2-7B-Instruct and Qwen2-72B-Instruct support extended context lengths up to 128K tokens, while smaller models adopt the Apache 2.0 license and the largest retains the Qianwen License.written, reviewed, cited

Qwen2-57B-A14B disables tie embedding because large sparse embeddings consume a large proportion of total model parameters.
Instruction-tuned models require the Needle in a Haystack task to verify correct long-context understanding beyond perplexity evaluation.
The Qwen2-72B-Instruct model performs comparably to GPT-4 in safety tests across four categories of multilingual unsafe queries.

Qwen2-Math

qwenlm.github.io2024-08-08officialauthorityHN front pageabout the Qwen2 familyHN 128 pts · 38 comments5,833 words

Qwen2-Math is a series of specialized math language models built upon the Qwen2 LLMs that significantly outperforms the mathematical capabilities of open-source models and even closed-source models. The largest model, Qwen2-Math-72B-Instruct, is evaluated on benchmarks including GSM8K, Math, and MMLU-STEM using few-shot chain-of-thought prompting. Instruction tuning combines a dense reward signal with a binary signal via Rejection Sampling and reinforcement learning with Group Relative Policy Optimization (GRPO) after SFT.written, reviewed, cited

The model series primarily supports English, with bilingual English and Chinese math models scheduled for release soon.
Evaluation of multi-choice benchmarks including MMLU STEM and GaoKao problems uses a 5-shot setting instead of zero-shot.
Performance reporting for complex competition evaluations includes Greedy, Maj@64, RM@64, and RM@256 metrics.

An Analysis of Chinese LLM Censorship and Bias with Qwen 2 Instruct

huggingface.co2024-06-09authorityabout the Qwen2 familyHN 16 pts · 6 comments3,368 words

Qwen 2 Instruct models operate under an Apache 2.0 license but exhibit unique government-mandated alignment with significantly (>80%) fewer refusals in Chinese than English. The deccp dataset documents these behaviors, showing that responses to sensitive topics often include defensive nationalist propaganda or warnings rather than standard refusals. While the models are competitive on benchmarks, their output varies wildly in tone depending on the language used.written, reviewed, cited

Qwen2-72B-Instruct ranks just below Llama-3-70b-Instruct on the ChatArena Leaderboard several days after release.
The deccp dataset contains approximately 95 refusal questions used to evaluate Qwen2 Instructs censorship behavior.

Analysis3

Alibaba Qwen2.5-Omni-7B: Open Source End-to-End Multimodal AI Model

alizila.com2025-03-27about the Qwen2 familyHN 29 pts · 3 comments

Qwen2.5-Max: The Next Leap in AI Innovation

medium.com2025-01-29about the Qwen2 familyHN 17 pts · 4 comments

Qwen2 72B, Gemini Flash, GPT-4, Opus Added to PVQ Leaderboard

pvq.app2024-06-12HN 4 pts · 0 comments

All coverage tracked for this model (8) — append-only, nothing is ever removed

Title	Source	Published	First seen	Signals
Alibaba Qwen2.5-Omni-7B: Open Source End-to-End Multimodal AI Model	alizila.com	2025-03-27	2026-10-07	hn
Qwen2.5-Max: The Next Leap in AI Innovation	medium.com	2025-01-29	2026-10-07	hn
Llama.cpp Now Supports Qwen2-VL (Vision Language Model)	github.com	2024-12-14	2026-10-07	hn
Qwen2-7B-Instruct with TensorRT-LLM: consistently high tokens/SEC	inferless.com	2024-09-05	2026-10-07	hn
Qwen2-Math	qwenlm.github.io official	2024-08-08	2026-10-07	hn
Qwen2 72B, Gemini Flash, GPT-4, Opus Added to PVQ Leaderboard	pvq.app	2024-06-12	2026-10-07	hn
An Analysis of Chinese LLM Censorship and Bias with Qwen 2 Instruct	huggingface.co	2024-06-09	2026-10-07	hn
Qwen2 LLM Released	qwenlm.github.io official	2024-06-06	2026-10-07	hn

Similar models

3 models

Similar models

Llama-3.3-70B-Instruct · 70.6B
same size band (XL), same task (text-generation), more providers
Qwen2.5-72B-Instruct · 72.7B
same size band (XL), same task (text-generation), more providers
Llama-3.1-70B-Instruct · 70.6B
same size band (XL), same task (text-generation)

Sources

3 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:42 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:46 UTC · Summary: 2026-10-07 15:15 UTC · Last verified 2026-10-25

Follow Qwen2-72B

Watch this model

/feeds/models/Qwen--Qwen2-72B.xml(RSS 2.0) · every change to Qwen2-72B
/feeds/families/Qwen--Qwen2.xml(RSS 2.0) · every change across the Qwen2 family

Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.