QwQ-32B

32.8B parameters, text in, text out, 40K tokens of context. 1 provider serves it.

Verified 2026-10-25 Rank #134 in LLMs · score 94.3 method v1 Sources 3 Downloads 30d 35.4k · #411 of 619 tracked

Changed this week: 2 new quantized builds on the Hub (168 in all).2026-10-24

Can I run it?20GB at INT4Estimatefits a 24 GB cardmethod

Can I use it?Apache 2.0Permissive; patent grantlicence file2026-10-24

Does it work?20.9on Aider Polyglot#8 of 13 measuredleaderboard2025-03-06

Do people use it?35.4kdownloads, 30 days#411 of 619 we track · below the median+1 likes, downloads −4% this week2,970 likesHugging Face2026-10-24

Worth reading

Picked by rule from 23 tracked pieces: the lab's own post, the most shared, a field report, the latest.

QwQ-32B-Preview sometimes switches between English and Chinese, which complicates human understanding of its outputs.

From Show HN: QwQ-32B APIs – o1 like reasoning at 1% the cost · news.ycombinator.com · 2025-01-15

The lab's own post

QwQ-32B: Embracing the Power of Reinforcement Learning

QwQ-32B is a 32 billion parameter model available under the Apache 2.0 license that matches DeepSeek-R1's performance despite having only 32 billion parameters versus 671 billion.

A code execution server assesses whether generated codes successfully pass predefined test cases during training.

qwenlm.github.io 2025-03-05 480 HN pts · 169 comments 735 words

Most discussed · 17 HN pts

Show HN: QwQ-32B APIs – o1 like reasoning at 1% the cost

Ubicloud launched inference APIs for the open source QwQ-32B-Preview model, which offers reasoning capabilities comparable to OpenAI's o1 at significantly lower cost.

QwQ-32B-Preview sometimes switches between English and Chinese, which complicates human understanding of its outputs.
The model lacks awareness of certain Python libraries because it was trained on a smaller dataset than o1.

news.ycombinator.com 2025-01-15 17 HN pts · 3 comments 710 words

Latest · 20 months ago

How to Run QwQ-32B effectively

Unsloth provides a guide for running QwQ-32B effectively, addressing issues like infinite generations and repetition.

The model requires manual removal of the <think> token from the Jinja chat template if systems fail to output thinking traces.
Unsloth uploaded dynamic 4-bit quants to Hugging Face that increase accuracy versus naive 4-bit quantizations.

docs.unsloth.ai 2025-03-07 4 HN pts · 3 comments 1k words

23 best of 23 tracked

The 3 picks above, plus these 20. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.

ArticleQwQ: Alibaba's O1-like reasoning LLMqwenlm.github.io · 2024-11-27 · 438 HN pts · official
ArticleLLM Comparison/Test: 25 SOTA LLMs (Including QwQ) Through 59 MMLU-Pro CS Runshuggingface.co · 2024-12-05 · 2 HN pts
ArticleShow HN: QwQ AI – Aggregator of free LLMs answer generatorqwq32.com · 2025-04-19 · 6 HN pts
CodeDon't Think Too Much (DTTM) Proxy for Qwen/QwQ-32Bgithub.com · 2025-03-10 · 1 HN pts
ArticleQwerky: Attention is not what you need? RWKV mashed into QwQ modelssubstack.recursal.ai · 2025-04-06 · 3 HN pts
CodeCombine answers from all reasoning models(deepseek-r1, o1, flash-thinking, qwq)github.com · 2025-01-25 · 1 HN pts
ArticleQwQ-32B: The Power of Scaling RL | QWQ AIqwq32.com · undated
ArticleI Tested QwQ-32B, Alibaba's New Reasoning AI – Here's Why It's Surprisingly Powerfullatenode.com · undated
ArticleQwQ 32B | Open Laboratoryopenlaboratory.com · undated
ArticleQwQ-32B — benchmarks, pricing & hardware fit · Model Beatsmodelbeats.com · undated
ArticleQwQ-32B — Edgepediaedgechat.ai · undated
ArticleDeepSeek-R1-beating perf 32B package? El Reg digs its claws into Alibaba's QwQtheregister.com · 2025-03-16
VideoQwQ: Tiny Thinking Model That Tops DeepSeek R1 (Open Source) [video]youtube.com · 2025-03-07
ArticleQwq – Deepseek but Betterollama.com · 2025-03-10
ArticleQwQ-32B Now Available on Featherless.aifeatherless.ai · 2025-03-07
ArticleQwQ-32B significantly better at competition problems than previous open modelsmathstodon.xyz · 2024-11-30
ArticleEverything to know about QwQ-32B, Alibaba's new reasoning modelbdtechtalks.substack.com · 2025-03-07
ArticleQwQ 32B: A Compact AI Rival to DeepSeek R1 - Novitablogs.novita.ai · 2025-03-07
ArticleQwQ-32B Reviewlearnwithlabio.com · 2025-03-06
ArticleQwQ-32B Review — Benchmarks, Pricing & AI Panel Verdict | TopReviewed.aitopreviewed.ai · 2025-03-04

Show all 20

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; heavily run locally (168 quantized builds); a big base for builders (219 fine-tunes, adapters and merges).

As a service1Featherless
On own hardware168quantized builds
Built on21994 fine-tunes · 69 adapters · 56 merges
In apps100+Hugging Face Spaces

Top Spaces running it: Qwen/QwQ-32B-Demo ♥ 560 · hadadrjt/ai ♥ 86 · modelscope/modelscope-studio ♥ 72 · yourbench/advanced ♥ 45

Will it fit?

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

Fits a 64 GB Mac (unified memory) at INT4

About 20 GB at INT4, 39 GB at FP8, 79 GB at BF16: weights only, so leave room for the KV cache.

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

How it scores

Third-party measurements only; author-reported numbers stay in Reference, marked as such.

#8 of 13 on Aider Polyglot so far

20.9. Each tick is a model on this site with a measured Aider Polyglot row; author-reported numbers stay in Reference. Percent (pass_rate_2). edit_format=diff

Source · leaderboard · 2025-03-06

The story so far

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

Most coverage came in the first two weeks

Teal is the weights (2025-03-05), amber triangles are 9 dated articles (each links to it), grey ticks are repo commits.

What people built on it

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

387 derivatives on the Hub, led by 168 quantized

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

GGUF MaziyarPanahi/QwQ-32B-GGUF for llama.cpp, Ollama, LM Studio 3.5M downloads MLX lmstudio-community/QwQ-32B-MLX-8bit for Apple silicon 761.8k downloads AWQ Qwen/QwQ-32B-AWQ for vLLM, SGLang 1.3M downloads

All variants →

Reference

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

Benchmarks

2 rows

Benchmarks 2 rows

measured third-party leaderboard row (2)author-reported number from the model card or eval filesNo 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
measured Aider Polyglot · Write code with agents	20.9 percent (pass_rate_2)	edit_format=diff	leaderboard	2025-03-06
measured Aider Polyglot · Write code with agents	26.2 percent (pass_rate_2)	edit_format=architect	leaderboard	2025-03-07

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.

Write code with agentsmeasuredAider Polyglot 26.2
Reason through hard problemsstructuraldecision-model estimate p=1.00 chat template: <think>
Chat and follow instructionsstructuraltag: conversational chat template present
Call tools and APIsstructuralchat template: <tool_call>
Read long documentsstructuraldecision-model estimate p=0.96

Facts

22 fields

Listed in LLMs

Facts

QwQ-32B has 32.8B parameters. (safetensors metadata, observed 2026-10-24)
Context length is 40,960 tokens (40K). (config.json max_position_embeddings, observed 2026-10-24)
License: Apache 2.0 (friction: none). Permissive; patent grant. (license file, observed 2026-10-24)
Created 2025-03-05, last modified 2025-03-11. (HF API, observed 2026-10-24)

Pipeline tag	text-generation
Library	transformers
Architectures	Qwen2ForCausalLM
Model type	qwen2
Dtype	bfloat16
Params by dtype	BF16: 32,763,876,352
Size band	L
MoE	no
Context length	40,960 (config.json max_position_embeddings)
Languages	en
Papers	arXiv:2309.00071, arXiv:2412.15115
License	Apache 2.0 (apache-2.0)
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 QwQ-32B: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)	79 GB · Single high-end GPU (<80 GB)
FP8 (1 byte/param)	39 GB · Single high-end GPU (<80 GB)
INT4 (0.5 bytes/param)	20 GB · Single GPU (<24 GB)

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

All variants

40 variants

Lineage

Fine-tuned from Qwen/Qwen2.5-32B (same organization) Family: Qwen/QwQ 387 derivatives on the Hub

Base model: Qwen/Qwen2.5-32B

quantized: 168finetune: 94merge: 56adapter: 69

Top variants (40 shown, ranked by downloads)

Variant	Relation	Quant	Params	Band	License	Author	Likes	Downloads
MaziyarPanahi/QwQ-32B-GGUF	quantized	GGUF	—	—	—	MaziyarPanahi	4	3.5M
Qwen/QwQ-32B-AWQ	quantized	AWQ	32.8B	L	apache-2.0	Qwen	142	1.3M
lmstudio-community/QwQ-32B-MLX-8bit	quantized	MLX	9.22B	L	apache-2.0	lmstudio-community	0	761.8k
Qwen/QwQ-32B-GGUF	quantized	GGUF	—	—	apache-2.0	Qwen	222	440.9k
lmstudio-community/QwQ-32B-GGUF	quantized	GGUF	—	—	apache-2.0	lmstudio-community	46	373.7k
bartowski/Qwen_QwQ-32B-GGUF	quantized	GGUF	—	—	apache-2.0	bartowski	167	310.4k
unsloth/QwQ-32B-GGUF	quantized	GGUF	—	—	apache-2.0	unsloth	86	190.2k
cognition-ai/Kevin-32B	finetune	—	32.8B	L	—	cognition-ai	164	91.5k
unsloth/QwQ-32B-unsloth-bnb-4bit	quantized	-bnb	32.8B	L	apache-2.0	unsloth	47	81.4k
RedHatAI/QwQ-32B-FP8-dynamic	quantized	-FP8	32.8B	L	mit	RedHatAI	14	26.5k
huihui-ai/QwQ-32B-abliterated	finetune	—	32.8B	L	apache-2.0	huihui-ai	111	16.8k
unsloth/QwQ-32B	finetune	—	32.8B	L	apache-2.0	unsloth	17	20k
mradermacher/QwQ-32B-i1-GGUF	quantized	GGUF	—	—	apache-2.0	mradermacher	5	19.8k
trashpanda-org/QwQ-32B-Snowdrop-v0	merge	—	32.8B	L	—	trashpanda-org	104	13k
mlx-community/QwQ-32B-4bit	quantized	-4bit	32.8B	L	apache-2.0	mlx-community	38	15.2k
unsloth/QwQ-32B-bnb-4bit	quantized	-bnb	33.7B	L	apache-2.0	unsloth	4	16.7k
ArliAI/QwQ-32B-ArliAI-RpR-v1	finetune	—	32.8B	L	apache-2.0	ArliAI	76	9.4k
ArliAI/QwQ-32B-ArliAI-RpR-v4	finetune	—	32.8B	L	apache-2.0	ArliAI	44	8.8k
Alibaba-NLP/WebDancer-32B	finetune	—	32.8B	L	mit	Alibaba-NLP	57	7.7k
mlx-community/QwQ-32B-8bit	quantized	-8bit	32.8B	L	apache-2.0	mlx-community	14	9.7k
openfree/QwQ-32B-Q4_K_M-GGUF	quantized	GGUF	—	—	apache-2.0	openfree	98	4.1k
openfree/QwQ-32B-Q8_0-GGUF	quantized	Q8_0	—	—	apache-2.0	openfree	104	3.6k
rinna/qwq-bakeneko-32b	merge	—	32.8B	L	apache-2.0	rinna	21	7.7k
abeja/ABEJA-QwQ32b-Reasoning-Japanese-v1.0	finetune	—	32.8B	L	apache-2.0	abeja	15	7.7k
mmnga/QwQ-32B-gguf	quantized	gguf	—	—	apache-2.0	mmnga	1	7.6k
ArliAI/QwQ-32B-ArliAI-RpR-v1-GGUF	quantized	GGUF	—	—	apache-2.0	ArliAI	19	6.6k
kaitchup/QwQ-32B-AWQ-4bit	quantized	AWQ	32.8B	L	apache-2.0	kaitchup	3	7.4k
VIDraft/QwQ-R1984-32B	finetune	—	32.8B	L	apache-2.0	VIDraft	150	3
OPEA/QwQ-32B-int4-AutoRound-gptq-sym	quantized	int4	32.8B	L	apache-2.0	OPEA	4	7.2k
ModelCloud/QwQ-32B-gptqmodel-4bit-vortex-v1	quantized	gptq	32.8B	L	apache-2.0	ModelCloud	12	6.5k
hltcoe/Rank-K-32B	finetune	—	32.8B	L	mit	hltcoe	8	6.4k
Mungert/QwQ-32B-ArliAI-RpR-v4-GGUF	quantized	GGUF	—	—	apache-2.0	Mungert	3	6.7k
mradermacher/QwQ-32B-GGUF	quantized	GGUF	—	—	apache-2.0	mradermacher	3	6.2k
brittlewis12/QwQ-32B-GGUF	quantized	GGUF	—	—	apache-2.0	brittlewis12	0	5.8k
ArliAI/QwQ-32B-ArliAI-RpR-v3	finetune	—	32.8B	L	apache-2.0	ArliAI	41	3.4k
kaitchup/QwQ-32B-AutoRoundGPTQ-8bit	quantized	GPTQ	32.8B	L	apache-2.0	kaitchup	2	5.2k
Mungert/Kevin-32B-GGUF	quantized	GGUF	—	—	—	Mungert	3	4.7k
ospatch/QwQ-32B-INT8-W8A8	quantized	INT8	32.8B	L	apache-2.0	ospatch	5	4.1k
tensorblock/QwQ-32B-GGUF	quantized	GGUF	—	—	apache-2.0	tensorblock	4	3.7k
second-state/QwQ-32B-GGUF	quantized	GGUF	—	—	apache-2.0	second-state	0	3.6k

Repository changelog

19 commits

changed in the last 7 daysOther: 2026-10-24all changes

Repository changelog 19 commits

Date	Commit	SHA
2025-03-11	update tokenizer_config.json	976055f
2025-03-11	update tokenizer.json	11e5ea0
2025-03-10	update README & config	2d67d9d
2025-03-07	Update README.md	b6306e1
2025-03-06	update README	f28e641
2025-03-06	Add transformers as library! (#1)	bac0d01
2025-03-05	update readme	7c0a8dc
2025-03-05	update readme	2ebdfe8
2025-03-05	Update README.md	3ffd731
2025-03-05	Update README.md	c096839
2025-03-05	Update tokenizer_config.json	d76cadc
2025-03-05	Update README.md	d9c3c2f
2025-03-05	Update README.md	c4e7023
2025-03-05	Update README.md	3a8761b
2025-03-05	Update config.json	88e1a85
2025-03-05	Create README.md	20dea38
2025-03-05	Update config.json	915876a
2025-03-05	Upload folder using huggingface_hub	883729b
2025-03-05	initial commit	8538ab7

All providers

1 provider · 1 endpoint

Where to run 1 providers · 1 endpoints

Hosted endpoints observed for QwQ-32B, 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/QwQ-32B	—	—	—	—	—	—	—	live	2026-10-25

Articles and coverage

23 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.

Official4

QwQ: Alibaba's O1-like reasoning LLM

qwenlm.github.io2024-11-27officialauthorityHN front pageHN 438 pts · 421 comments2,950 words

QwQ-32B-Preview is an experimental research model developed by the Qwen Team focused on advancing AI reasoning capabilities. The model achieves 65.2% on GPQA, 50.0% on AIME, 90.6% on MATH-500, and 50.0% on LiveCodeBench.written, reviewed, cited

The model may enter circular reasoning patterns, leading to lengthy responses without a conclusive answer.
Users should exercise caution when deploying the model due to required enhanced safety measures for reliable performance.
The model may mix languages or switch between them unexpectedly, affecting response clarity.

QwQ-32B: Embracing the Power of Reinforcement Learning

qwenlm.github.io2025-03-05officialauthorityHN front pageHN 480 pts · 169 comments735 words

QwQ-32B is a 32 billion parameter model available under the Apache 2.0 license that matches DeepSeek-R1's performance despite having only 32 billion parameters versus 671 billion. It employs an accuracy verifier for math and a code execution server during initial reinforcement learning stages before adding general capability training. Users can generate responses with a maximum of 32768 new tokens via Hugging Face or Alibaba Cloud DashScope API.written, reviewed, cited

A code execution server assesses whether generated codes successfully pass predefined test cases during training.

How to Run QwQ-32B effectively

docs.unsloth.ai2025-03-07authorityHN 4 pts · 3 comments1,018 words

Unsloth provides a guide for running QwQ-32B effectively, addressing issues like infinite generations and repetition. The guide recommends a Temperature of 0.6, Top_K of 40, and a Repetition Penalty of 1.0 to ensure stable performance. For llama.cpp, users must reorder samplers to top_k;top_p;min_p;temperature;dry;typ_p;xtc before applying the penalty to prevent endless generations.written, reviewed, cited

The model requires manual removal of the <think> token from the Jinja chat template if systems fail to output thinking traces.
Unsloth uploaded dynamic 4-bit quants to Hugging Face that increase accuracy versus naive 4-bit quantizations.

LLM Comparison/Test: 25 SOTA LLMs (Including QwQ) Through 59 MMLU-Pro CS Runs

huggingface.co2024-12-05authorityabout the QwQ familyHN 2 pts · 0 comments5,222 words

Claude 3.5 Sonnet achieved an 82.93% score on the 410-question MMLU-Pro computer science benchmark, ranking as the top performer. QwQ 32B Preview secured the best local model position with a 79.27% score, outperforming larger 70B and 405B alternatives. This 32B model requires max_tokens=16K to fully utilize its chain-of-thought reasoning capabilities and performs best in the 8.0bpw EXL2 quantization format.written, reviewed, cited

The MMLU-Pro benchmark uses 10 answer options per question instead of 4 to reduce chance correctness.
Testing 25 models required over 70 hours of runtime, limiting the study to the Computer Science category.
QwQ 32B Preview performs worse if output is truncated because it needs max_tokens=16K to think.

Analysis5

Show HN: QwQ AI – Aggregator of free LLMs answer generator

qwq32.com2025-04-19about the QwQ familyHN 6 pts · 1 comments455 words

Don't Think Too Much (DTTM) Proxy for Qwen/QwQ-32B

github.com2025-03-10authorityHN 1 pts · 1 comments417 words

Qwerky: Attention is not what you need? RWKV mashed into QwQ models

substack.recursal.ai2025-04-06about the QwQ familyHN 3 pts · 1 comments672 words

Combine answers from all reasoning models(deepseek-r1, o1, flash-thinking, qwq)

github.com2025-01-25authorityabout the QwQ familyHN 1 pts · 0 comments581 words

QwQ-32B: The Power of Scaling RL | QWQ AI

qwq32.com

Community1

Show HN: QwQ-32B APIs – o1 like reasoning at 1% the cost

news.ycombinator.com2025-01-15HN 17 pts · 3 comments710 words

Ubicloud launched inference APIs for the open source QwQ-32B-Preview model, which offers reasoning capabilities comparable to OpenAI's o1 at significantly lower cost. The service prices API endpoints at $0.60 per M tokens, which is 100x lower than o1's output token price, and offers the first million tokens each month for free. In the MMLU-Pro Computer Science LLM Benchmark, QwQ-32B-Preview achieves 79.1% accuracy, ranking behind Claude-3.5 Sonnet (82.5) but ahead of GPT 4o 2024-11-20 (73.1).written, reviewed, cited

QwQ-32B-Preview sometimes switches between English and Chinese, which complicates human understanding of its outputs.
The model lacks awareness of certain Python libraries because it was trained on a smaller dataset than o1.
QwQ-32B-Preview enumerates answers using light heuristics while o1 employs a deeper iterative deepen-and-test strategy.

All coverage tracked for this model (23) — append-only, nothing is ever removed

Title	Source	Published	First seen	Signals
QwQ-32B: The Power of Scaling RL \| QWQ AI	qwq32.com	—	2026-10-07	exa
I Tested QwQ-32B, Alibaba's New Reasoning AI – Here's Why It's Surprisingly Powerful	latenode.com	—	2026-10-07	exa
QwQ 32B \| Open Laboratory	openlaboratory.com	—	2026-10-07	exa
QwQ-32B — benchmarks, pricing & hardware fit · Model Beats	modelbeats.com	—	2026-10-07	exa
QwQ-32B — Edgepedia	edgechat.ai	—	2026-10-07	exa
Show HN: QwQ AI – Aggregator of free LLMs answer generator	qwq32.com	2025-04-19	2026-10-07	hn
Qwerky: Attention is not what you need? RWKV mashed into QwQ models	substack.recursal.ai	2025-04-06	2026-10-07	hn
DeepSeek-R1-beating perf 32B package? El Reg digs its claws into Alibaba's QwQ	theregister.com	2025-03-16	2026-10-07	hn
Qwq – Deepseek but Better	ollama.com	2025-03-10	2026-10-07	hn
Don't Think Too Much (DTTM) Proxy for Qwen/QwQ-32B	github.com	2025-03-10	2026-10-07	hn
QwQ: Tiny Thinking Model That Tops DeepSeek R1 (Open Source) [video]	youtube.com	2025-03-07	2026-10-07	hn
How to Run QwQ-32B effectively	docs.unsloth.ai	2025-03-07	2026-10-07	hn
QwQ-32B Now Available on Featherless.ai	featherless.ai	2025-03-07	2026-10-07	hn
Everything to know about QwQ-32B, Alibaba's new reasoning model	bdtechtalks.substack.com	2025-03-07	2026-10-07	exa
QwQ 32B: A Compact AI Rival to DeepSeek R1 - Novita	blogs.novita.ai	2025-03-07	2026-10-07	exa
QwQ-32B Review	learnwithlabio.com	2025-03-06	2026-10-07	exa
QwQ-32B: Embracing the Power of Reinforcement Learning	qwenlm.github.io official	2025-03-05	2026-10-07	hn
QwQ-32B Review — Benchmarks, Pricing & AI Panel Verdict \| TopReviewed.ai	topreviewed.ai	2025-03-04	2026-10-07	exa
Combine answers from all reasoning models(deepseek-r1, o1, flash-thinking, qwq)	github.com	2025-01-25	2026-10-07	hn
Show HN: QwQ-32B APIs – o1 like reasoning at 1% the cost	news.ycombinator.com	2025-01-15	2026-10-07	hn
LLM Comparison/Test: 25 SOTA LLMs (Including QwQ) Through 59 MMLU-Pro CS Runs	huggingface.co	2024-12-05	2026-10-07	hn
QwQ-32B significantly better at competition problems than previous open models	mathstodon.xyz	2024-11-30	2026-10-07	hn
QwQ: Alibaba's O1-like reasoning LLM	qwenlm.github.io official	2024-11-27	2026-10-07	hn

Papers

2 papers

Papers 2

arXiv links found in the model card and its tags. "Peer-reviewed" is shown only with evidence: a published venue in OpenAlex (measured) or the authors' own journal reference on arXiv (reported).

Qwen2.5 Technical Report cited in card

Qwen, An Yang, Baosong Yang +40 · 2024-12-19 · PDF · HF paper page (381 upvotes)
YaRN: Efficient Context Window Extension of Large Language Models cited in card

Bowen Peng, Jeffrey Quesnelle, Honglu Fan +1 · 2023-08-31 · PDF · HF paper page (88 upvotes)

Similar models

3 models

Similar models

Qwen3-32B · 32.8B
same size band (L), same task (text-generation), more providers, permissive license
Qwen2.5-Coder-32B-Instruct · 32.8B
same size band (L), same task (text-generation), more providers, permissive license
DeepSeek-R1-Distill-Qwen-32B · 32.8B
same size band (L), same task (text-generation), permissive license

Sources

3 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:34 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 15:01 UTC · Last verified 2026-10-25

Follow QwQ-32B

Watch this model

/feeds/models/Qwen--QwQ-32B.xml(RSS 2.0) · every change to QwQ-32B
/feeds/families/Qwen--QwQ.xml(RSS 2.0) · every change across the QwQ family

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