Qwen3-32B

32.8B parameters, text in, text out, 40K tokens of context. 5 providers serve it from $0.080 per million input tokens.

Verified 2026-10-25 Rank #31 in LLMs · score 113.3 method v1 Sources 4 Downloads 30d 2.9M · #51 of 619 tracked

Changed this week: 8 new quantized builds on the Hub (188 in all).2026-10-24 · 1 new provider: Nscale.2026-10-23

Can I run it?20GB at INT4Estimatefits a 24 GB cardmethodCommunity: Qwen/Qwen3-32B Hardware, VRAM, and Deployment Guide | InnoAI →

What does it cost?$0.080/1M incheapest of 4Nscale2026-10-25

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

Does it work?48.71on BFCL#5 of 28 measuredleaderboard2026-10-25

Do people use it?2.9Mdownloads, 30 days#51 of 619 we track · top 8%+4 likes, downloads −5% this week772 likesHugging Face2026-10-24

Worth reading

Picked by rule from 19 tracked pieces: the lab's own post, the most shared, a field report, the latest.

Most discussed · 490 HN pts

GPT-OSS vs. Qwen3 and a detailed look how things evolved since GPT-2

OpenAI released gpt-oss-120b and gpt-oss-20b, their first open-weight models since GPT-2 in 2019.

A 2025 research paper confirms Dropout results in worse downstream performance in single-epoch regimes.

magazine.sebastianraschka.com 2025-08-10 490 HN pts · 97 comments 5.2k words about the Qwen3 family family

Latest · 8 months ago

Qwen3.5 122B and 35B models offer Sonnet 4.5 performance on local computers

venturebeat.com 2026-02-28 461 HN pts · 272 comments about the Qwen3 family family

19 best of 19 tracked

The 4 picks above, plus these 15. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.

ArticleQwen3-Coder-Nextqwen.ai · 2026-02-03 · official
ArticleQwen3-Max-Thinkingqwen.ai · 2026-01-26 · official
ArticleQwen3-TTS family is now open sourced: Voice design, clone, and generationqwen.ai · 2026-01-22 · official
ArticleQwen3: Think deeper, act fasterqwenlm.github.io · 2025-04-28 · 869 HN pts · official
ArticleQwen3-Coder: Agentic coding in the worldqwenlm.github.io · 2025-07-22 · official
ArticleQwen3-Nextqwen.ai · 2025-09-12 · official
ArticleQwen3-VLqwen.ai · 2025-09-23 · official
CodeQwen3-Omni: Native Omni AI model for text, image and videogithub.com · 2025-09-22
PaperQwen3 Technical Report - arXivarxiv.org · undated
ArticleQwen3 32B (non-reasoning) - Artificial Analysisartificialanalysis.ai · undated
ArticleQwen3: Features, DeepSeek-R1 Comparison, Access, and Moredatacamp.com · undated
ArticleQwen3 32B - 32B model, AA Index 8.6 | Every Local AIeverylocalai.com · undated
ArticleQwen/Qwen3-32B Hardware, VRAM, and Deployment Guide | InnoAIinnoai.space · undated
ArticleQwen3 Released: How Does It Stack Up?runpod.io · undated
ArticleQwen3-32B Review — Benchmarks, Pricing & AI Panel Verdict | TopReviewed.aitopreviewed.ai · 2025-04-28

Show all 15

Where to run it

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

Prices for the same weights vary 2×

4 providers with a live price, from $0.080 to $0.14 per million input tokens; the median is $0.11. One dot per provider, cheapest in teal.

#	Cheapest 5	In /1M	Out /1M	Context	Quant	Seen
1	Nscale	$0.080	$0.25	40K	—	Oct 25
2	DeepInfra	$0.080	$0.28	40K	fp8	Oct 25
3	Tenstorrent	$0.14	$0.40	20K	fp8	Oct 24
4	SiliconFlow	$0.14	$0.57	128K	fp8	Oct 24
5	Featherless	—	—	—	—	Oct 25

All 5 providers →

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.

Widely served (5 API providers); heavily run locally (188 quantized builds); a big base for builders (1,100 fine-tunes, adapters and merges).

As a service5DeepInfra, Featherless, Nscale, SiliconFlow +1
On own hardware188quantized builds
Built on1,100593 fine-tunes · 488 adapters · 19 merges
In apps100+Hugging Face Spaces

Top Spaces running it: Qwen/Qwen3-Demo ♥ 868 · pliny-the-prompter/obliteratus ♥ 621 · embedl/hfviewer ♥ 181 · fdaudens/podcast-jobs ♥ 81

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.

#5 of 28 on BFCL so far

48.71. Each tick is a model on this site with a measured BFCL row; author-reported numbers stay in Reference. Percent (overall accuracy). FC

Source · leaderboard · 2026-10-25

The story so far

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

3 dated articles and 19 repo commits so far

Teal is the weights (2025-04-27), amber triangles are 3 dated articles (each links to it), grey ticks are repo commits.

What people built on it

1,288 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.

1,288 derivatives on the Hub, led by 593 fine-tunes

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

GGUF MaziyarPanahi/Qwen3-32B-GGUF for llama.cpp, Ollama, LM Studio 3.4M downloads MLX lmstudio-community/Qwen3-32B-MLX-4bit for Apple silicon 541.2k downloads AWQ Qwen/Qwen3-32B-AWQ for vLLM, SGLang 10.1M downloads

All variants →

Reference

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

Benchmarks

3 rows

Benchmarks 3 rows

measured third-party leaderboard row (3)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	40 percent (pass_rate_2)	edit_format=diff	leaderboard	2025-05-08
measured BFCL · Call tools and APIs	48.71 percent (overall accuracy)	FC 5 sub-scores Non-Live AST Acc: 88.77 Live Acc: 82.01 Multi Turn Acc: 47.87 Web Search Acc: 21.5 Memory Acc: 26.67	leaderboard	2026-10-25
measured BFCL · Call tools and APIs	46.78 percent (overall accuracy)	Prompt 5 sub-scores Non-Live AST Acc: 90.27 Live Acc: 82.01 Multi Turn Acc: 43.25 Web Search Acc: 26 Memory Acc: 15.7	leaderboard	2026-10-25

What the card says

1 cited sentence

Summary generated, cited

The architecture features 64 layers and supports a native context length of 32,768 tokens.^[1]^[2]

“Number of Layers: 64”
“Context Length: 32,768 natively”

Generated 2026-10-07 14:59 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.

Facts

21 fields

Listed in LLMs

Facts

Qwen3-32B 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 2025-04-27, last modified 2025-07-26. (HF API, observed 2026-10-24)

Pipeline tag	text-generation
Library	transformers
Architectures	Qwen3ForCausalLM
Model type	qwen3
Dtype	bfloat16
Params by dtype	BF16: 32,762,123,264
Size band	L
MoE	no
Context length	40,960 (config.json max_position_embeddings)
Papers	arXiv:2309.00071, arXiv:2505.09388
License	Apache 2.0 (apache-2.0)
Hub snapshot	2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: supported · Qwen3ForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · Qwen3ForCausalLM · 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 Qwen3-32B: 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

Original model Family: Qwen/Qwen3 1,288 derivatives on the Hub

quantized: 188merge: 19finetune: 593adapter: 488

Top variants (40 shown, ranked by downloads)

Variant	Relation	Quant	Params	Band	License	Author	Likes	Downloads
Qwen/Qwen3-32B-AWQ	quantized	AWQ	32.8B	L	apache-2.0	Qwen	151	10.1M
Qwen/Qwen3-32B-FP8	quantized	-FP8	32.8B	L	apache-2.0	Qwen	92	4.5M
MaziyarPanahi/Qwen3-32B-GGUF	quantized	GGUF	—	—	—	MaziyarPanahi	2	3.4M
unsloth/Qwen3-32B-bnb-4bit	quantized	-bnb	33.7B	L	—	unsloth	7	827k
unsloth/Qwen3-32B-GGUF	quantized	GGUF	—	—	apache-2.0	unsloth	101	789.6k
RedHatAI/Qwen3-32B-NVFP4	quantized	NVFP4	19.1B	L	apache-2.0	RedHatAI	9	725.3k
nvidia/Qwen3-32B-NVFP4	quantized	NVFP4	17.2B	L	apache-2.0	nvidia	26	704.9k
Qwen/Qwen3-32B-GGUF	quantized	GGUF	—	—	apache-2.0	Qwen	76	667.3k
lmstudio-community/Qwen3-32B-MLX-4bit	quantized	MLX	32.8B	L	apache-2.0	lmstudio-community	7	541.2k
lmstudio-community/Qwen3-32B-MLX-8bit	quantized	MLX	32.8B	L	apache-2.0	lmstudio-community	2	524.6k
LLMcompe-Team-Watanabe/Qwen3-32B-merge-base2-math3-science3-submath05-med05-other1	merge	—	32.8B	L	—	LLMcompe-Team-Watanabe	0	384k
Goedel-LM/Goedel-Prover-V2-32B	finetune	—	32.8B	L	apache-2.0	Goedel-LM	76	316.1k
unsloth/Qwen3-32B	finetune	—	32.8B	L	—	unsloth	15	232.1k
lmstudio-community/Qwen3-32B-GGUF	quantized	GGUF	—	—	apache-2.0	lmstudio-community	11	218.8k
unsloth/Qwen3-32B-unsloth-bnb-4bit	quantized	-bnb	33.3B	L	apache-2.0	unsloth	15	200.8k
RedHatAI/Qwen3-32B-FP8-dynamic	quantized	-FP8	32.8B	L	apache-2.0	RedHatAI	15	192.3k
bartowski/Qwen_Qwen3-32B-GGUF	quantized	GGUF	—	—	apache-2.0	bartowski	41	157.5k
microsoft/FrogBoss-32B-2510	finetune	—	677K	S	mit	microsoft	44	111.2k
JunHowie/Qwen3-32B-GPTQ-Int4	quantized	GPTQ	32.8B	L	apache-2.0	JunHowie	4	110.7k
unsloth/Qwen3-32B-128K-GGUF	quantized	GGUF	—	—	apache-2.0	unsloth	27	78.1k
RedHatAI/Qwen3-32B-quantized.w4a16	quantized	-quant	32.8B	L	apache-2.0	RedHatAI	12	76.4k
amazon/GKA-primed-HQwen3-32B-Instruct	finetune	—	34.4B	L	apache-2.0	amazon	2	65.9k
yujiepan/qwen3-tiny-random-tp	finetune	—	2.44M	S	—	yujiepan	0	51k
t-tech/T-pro-it-2.0	finetune	—	32.8B	L	apache-2.0	t-tech	130	43.6k
agentica-org/DeepSWE-Preview	finetune	—	32.8B	L	mit	agentica-org	204	38.3k
geoffmunn/Qwen3-32B-f16	quantized	GGUF	—	—	apache-2.0	geoffmunn	2	41.2k
RedHatAI/Qwen3-32B-NVFP4A16	quantized	—	19.1B	L	apache-2.0	RedHatAI	2	37.4k
huihui-ai/Qwen3-32B-abliterated	finetune	—	32.8B	L	apache-2.0	huihui-ai	45	31.4k
6block/Qwen3-32B-GGUF	quantized	GGUF	—	—	apache-2.0	6block	1	26.1k
aisingapore/Qwen-SEA-LION-v4-32B-IT	finetune	—	32.8B	L	mit	aisingapore	8	23k
Sunbird/Sunflower-32B	finetune	—	32.8B	L	apache-2.0	Sunbird	2	22.8k
mlx-community/Qwen3-32B-4bit	quantized	-4bit	32.8B	L	apache-2.0	mlx-community	7	22.3k
biomni/Biomni-R0-32B-Preview	finetune	—	32.8B	L	mit	biomni	31	20.3k
amazon/GKA-primed-HQwen3-32B-Reasoner	finetune	—	34.1B	L	apache-2.0	amazon	4	21.2k
johnbridges/UIGEN-X-32B-0727-GGUF	quantized	GGUF	—	—	apache-2.0	johnbridges	0	21.3k
t-tech/T-pro-it-2.1	finetune	—	32.8B	L	apache-2.0	t-tech	36	18.1k
open-thoughts/OpenThinkerAgent-32B	finetune	—	677K	S	apache-2.0	open-thoughts	11	18.7k
RefalMachine/RuadaptQwen3-32B-Instruct-GGUF	quantized	GGUF	—	—	apache-2.0	RefalMachine	15	18.4k
second-state/Qwen3-32B-GGUF	quantized	GGUF	—	—	other	second-state	13	15.7k
DavidAU/Qwen3-32B-128k-NEO-Imatrix-Max-GGUF	quantized	Imatrix	—	—	apache-2.0	DavidAU	6	14.2k

Repository changelog

19 commits

changed in the last 7 daysNew providers: 2026-10-23Other: 2026-10-24all changes

Repository changelog 19 commits

Date	Commit	SHA
2025-07-26	Create LICENSE	9216db5
2025-05-21	Update README.md	d47b0d4
2025-05-19	update tokenizer_config.json	a09d5cc
2025-04-29	Update README.md	30b8421
2025-04-28	Update README.md	ba1f828
2025-04-28	Update README.md	c37ea5f
2025-04-28	Update README.md	c36b753
2025-04-28	Delete special_tokens_map.json	be2f114
2025-04-28	Delete added_tokens.json	e3a3f59
2025-04-28	Update README.md	4d178d9
2025-04-28	Update README.md	395feeb
2025-04-28	Update README.md	b1f6143
2025-04-28	Update README.md	c141ee0
2025-04-28	Update README.md	817577e
2025-04-28	Update README.md	940b12c
2025-04-28	Update README.md	c6740cd
2025-04-28	Update README.md	88f1f1f
2025-04-28	Upload folder using huggingface_hub	1aecd22
2025-04-27	initial commit	4c0e0ce

All providers

5 providers · 5 endpoints

Where to run 5 providers · 5 endpoints

Hosted endpoints observed for Qwen3-32B, sorted by input price (USD per million tokens). Each row names its observer and date; prices change daily, so check the source before quoting. Cheapest observed input price: $0.08/M.

Provider	Model ref	Context	Quant	$/M in	$/M out	tok/s	TTFT	Tools	Status	Observed
Nscale hf	Qwen/Qwen3-32B	40K	—	$0.08	$0.25	45.9	421 ms	yes	live	2026-10-25
DeepInfra hf, deepinfra, openrouter	Qwen/Qwen3-32B	40K	fp8	$0.08	$0.28	32.2	611 ms	yes	live	2026-10-25
Tenstorrent openrouter	qwen/qwen3-32b	20K	fp8	$0.14	$0.40	—	—	yes	live	2026-10-24
SiliconFlow openrouter	qwen/qwen3-32b	128K	fp8	$0.14	$0.57	—	—	yes	live	2026-10-24
Featherless hf	Qwen/Qwen3-32B	—	—	—	—	—	—	—	live	2026-10-25

Articles and coverage

19 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.

Official1

Qwen3: Think deeper, act faster

qwenlm.github.io2025-04-28officialauthorityHN front pageabout the Qwen3 familyHN 869 pts · 388 comments2,139 words

Qwen3 introduces Hybrid Thinking Modes, allowing users to toggle between step-by-step reasoning and rapid responses via computational budget control. The open-weight MoE model Qwen3-235B-A22B achieves performance similar to dense models while using only 10% of the active parameters. Pre-training utilized approximately 36 trillion tokens across 119 languages, with deployment supported by SGLang and vLLM.written, reviewed, cited

Qwen3-8B and Qwen3-14B models do not use tied embeddings, unlike smaller variants.
The pre-training pipeline includes a final stage extending context length to 32K tokens.

Analysis9

Qwen3.5 122B and 35B models offer Sonnet 4.5 performance on local computers

venturebeat.com2026-02-28HN front pageabout the Qwen3 familyHN 461 pts · 272 comments

GPT-OSS vs. Qwen3 and a detailed look how things evolved since GPT-2

magazine.sebastianraschka.com2025-08-10authorityHN front pageabout the Qwen3 familyHN 490 pts · 97 comments5,195 words

OpenAI released gpt-oss-120b and gpt-oss-20b, their first open-weight models since GPT-2 in 2019. The 20B model can run on a consumer GPU with up to 16 GB of RAM, and the 120B model can run on a single H100 with 80 GB of RAM or newer hardware. The models utilize MXFP4 optimization to fit onto single GPUs.written, reviewed, cited

A 2025 research paper confirms Dropout results in worse downstream performance in single-epoch regimes.

Qwen3 Technical Report - arXiv

arxiv.orgauthorityabout the Qwen3 family

Qwen3 32B (non-reasoning) - Artificial Analysis

artificialanalysis.aiauthority

Qwen 3 32B Review: 30 Days On Local Hardware | BestLLMfor

bestllmfor.com

Qwen3 32B Benchmarks, Pricing & Context Window - LLM Stats

llm-stats.com

Qwen3: Features, DeepSeek-R1 Comparison, Access, and More

datacamp.comabout the Qwen3 family

Qwen3 32B - 32B model, AA Index 8.6 | Every Local AI

everylocalai.com

Qwen/Qwen3-32B Hardware, VRAM, and Deployment Guide | InnoAI

innoai.space

All coverage tracked for this model (19) — append-only, nothing is ever removed

Title	Source	Published	First seen	Signals
Qwen3 Technical Report - arXiv	arxiv.org	—	2026-10-07	exa
Qwen3 32B (non-reasoning) - Artificial Analysis	artificialanalysis.ai	—	2026-10-07	exa
Qwen 3 32B Review: 30 Days On Local Hardware \| BestLLMfor	bestllmfor.com	—	2026-10-07	exa
Qwen3 32B Benchmarks, Pricing & Context Window - LLM Stats	llm-stats.com	—	2026-10-07	exa
Qwen3: Features, DeepSeek-R1 Comparison, Access, and More	datacamp.com	—	2026-10-07	exa
Qwen3 32B - 32B model, AA Index 8.6 \| Every Local AI	everylocalai.com	—	2026-10-07	exa
Qwen/Qwen3-32B Hardware, VRAM, and Deployment Guide \| InnoAI	innoai.space	—	2026-10-07	exa
Qwen3 Released: How Does It Stack Up?	runpod.io	—	2026-10-07	exa
Qwen3.5 122B and 35B models offer Sonnet 4.5 performance on local computers	venturebeat.com	2026-02-28	2026-10-07	hn
Qwen3-Coder-Next	qwen.ai official	2026-02-03	2026-10-07	hn
Qwen3-Max-Thinking	qwen.ai official	2026-01-26	2026-10-07	hn
Qwen3-TTS family is now open sourced: Voice design, clone, and generation	qwen.ai official	2026-01-22	2026-10-07	hn
Qwen3-VL	qwen.ai official	2025-09-23	2026-10-07	hn
Qwen3-Omni: Native Omni AI model for text, image and video	github.com	2025-09-22	2026-10-07	hn
Qwen3-Next	qwen.ai official	2025-09-12	2026-10-07	hn
GPT-OSS vs. Qwen3 and a detailed look how things evolved since GPT-2	magazine.sebastianraschka.com	2025-08-10	2026-10-07	hn
Qwen3-Coder: Agentic coding in the world	qwenlm.github.io official	2025-07-22	2026-10-07	hn
Qwen3: Think deeper, act faster	qwenlm.github.io official	2025-04-28	2026-10-07	hn, exa
Qwen3-32B Review — Benchmarks, Pricing & AI Panel Verdict \| TopReviewed.ai	topreviewed.ai	2025-04-28	2026-10-07	exa

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).

Qwen3 Technical Report this model

An Yang, Anfeng Li, Baosong Yang +57 · 2025-05-14 · PDF · HF paper page (349 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

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

Sources

4 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:32 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 14:59 UTC · Last verified 2026-10-25

Follow Qwen3-32B

Watch this model

/feeds/models/Qwen--Qwen3-32B.xml(RSS 2.0) · every change to Qwen3-32B
/feeds/families/Qwen--Qwen3.xml(RSS 2.0) · every change across the Qwen3 family

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