Model Follower.

Up to date

Rebuilt
2026-10-25 06:54 UTC
Release radar
117 on watch
Hub snapshot
2026-10-25
Re-checked
daily · every number sourced and dated

Models › Qwen3 family

Qwen3-30B-A3B-Base

Hugging Face ↗ Follow changes

Qwen Apache 2.0 out 18 months ago

30.5B parameters, 3B active (8 of 128 experts per token), text in, text out.

Verified 2026-10-25 Rank #722 in LLMs · score 71.7 method v1 Sources 3 Downloads 30d 36.9k · #405 of 619 tracked

Can I run it?18GB at INT4Estimatefits a 24 GB cardmethod
Can I use it?Apache 2.0Permissive; patent grantlicence file
Do people use it?36.9kdownloads, 30 days#405 of 619 we track · below the median+2 likes, downloads −9% this week86 likesHugging Face

Worth reading

Picked by rule from 10 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

10 best of 10 tracked

The 2 picks above, plus these 8. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.

  1. ArticleQwen3-Coder-Nextqwen.ai · 2026-02-03 · official
  2. ArticleQwen3-Max-Thinkingqwen.ai · 2026-01-26 · official
  3. ArticleQwen3-TTS family is now open sourced: Voice design, clone, and generationqwen.ai · 2026-01-22 · official
  4. ArticleQwen3: Think deeper, act fasterqwenlm.github.io · 2025-04-28 · 869 HN pts · official
  5. ArticleQwen3-Coder: Agentic coding in the worldqwenlm.github.io · 2025-07-22 · official
  6. ArticleQwen3-Nextqwen.ai · 2025-09-12 · official
  7. ArticleQwen3-VLqwen.ai · 2025-09-23 · official
  8. CodeQwen3-Omni: Native Omni AI model for text, image and videogithub.com · 2025-09-22

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.

No API provider serves it, so its downloads are people running it themselves: 28 quantized builds.

  • As a service0no API provider listed
  • On own hardware28quantized builds
  • Built on11681 fine-tunes · 24 adapters · 11 merges
  • In apps0Hugging Face Spaces

Will it fit?

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

Fits a 64 GB Mac (unified memory) at INT4

About 18 GB at INT4, 37 GB at FP8, 73 GB at BF16: weights only, so leave room for the KV cache. Every expert must be resident, so total parameters count.

16 GB card64 GB Mac2×80 GB8×80 GBINT4: about 18 GB of weights (estimate; KV cache and activations extra)INT418 GBFP8: about 37 GB of weights (estimate; KV cache and activations extra)FP837 GBBF16: about 73 GB of weights (estimate; KV cache and activations extra)BF1673 GB
16 GB card64 GB Mac2×80 GB8×80 GBINT4: about 18 GB of weights (estimate; KV cache and activations extra)INT418 GBFP8: about 37 GB of weights (estimate; KV cache and activations extra)FP837 GBBF16: about 73 GB of weights (estimate; KV cache and activations extra)BF1673 GB

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

The story so far

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

3 dated articles and 0 repo commits so far

Teal is the weights (2025-04-28), amber triangles are 3 dated articles (each links to it).

MayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctQwen3.5 122B and 35B models offer Sonnet 4.5 performance on local computers · 2026-02-28Qwen3: Think deeper, act faster · 2025-04-28GPT-OSS vs. Qwen3 and a detailed look how things evolved since GPT-2 · 2025-08-10Weights 2025-04-28 · 2025-04-28Weights 2025-04-28now
MayJunJulAugSepOctNovDecJanFebMarAprMayJunJulAugSepOctQwen3.5 122B and 35B models offer Sonnet 4.5 performance on local computers · 2026-02-28Qwen3: Think deeper, act faster · 2025-04-28GPT-OSS vs. Qwen3 and a detailed look how things evolved since GPT-2 · 2025-08-10Weights 2025-04-28 · 2025-04-28Weights 2025-04-28now

What people built on it

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

144 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.

28 quantized (Hugging Face model tree)81 fine-tunes (Hugging Face model tree)11 merges (Hugging Face model tree)24 adapters (Hugging Face model tree)28 quantized81 fine-tunes11 merges24 adapters
28 quantized (Hugging Face model tree)81 fine-tunes (Hugging Face model tree)11 merges (Hugging Face model tree)24 adapters (Hugging Face model tree)28 quantized81 fine-tunes11 merges24 adapters
GGUF Mungert/Qwen3-30B-A6B-16-Extreme-GGUF for llama.cpp, Ollama, LM Studio 8.9k downloads GPTQ AlphaGaO/Qwen3-30B-A3B-GPTQ for vLLM, SGLang 8.2k downloads

All variants →

Reference

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

Use-case evidence

1 use case

What it is good at 1 use case

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

17 fields

Listed in LLMs · MoE

Facts

Pipeline tagtext-generation
Librarytransformers
ArchitecturesQwen3MoeForCausalLM
Model typeqwen3_moe
Params by dtypeBF16: 30,532,122,624
Size bandL
MoEyes
Experts128
Experts per token8
PapersarXiv:2505.09388
LicenseApache 2.0 (apache-2.0)
Hub snapshot2026-10-25

Hardware fit

3 estimates

Hardware fit ESTIMATE

Weights-only memory for Qwen3-30B-A3B-Base: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)73 GB · Single high-end GPU (<80 GB)
FP8 (1 byte/param)37 GB · Single high-end GPU (<80 GB)
INT4 (0.5 bytes/param)18 GB · Single GPU (<24 GB)

Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra. MoE: all experts must be resident, so total parameters count.

All variants

40 variants

Lineage

Original model Family: Qwen/Qwen3 144 derivatives on the Hub

finetune: 81quantized: 28merge: 11adapter: 24

Top variants (40 shown, ranked by downloads)

Relation
Quant
Size
40 rows
Variant Relation Quant Params Band License Author Likes Downloads
unsloth/Qwen3-30B-A3B-Base finetune — 30.5B L apache-2.0 unsloth 4 20.4k
Jackmin108/Qwen3-30B-A3B-Fast finetune — 30.5B L apache-2.0 Jackmin108 0 15.9k
unsloth/Qwen3-30B-A3B-Base-bnb-4bit quantized -bnb 31.5B L — unsloth 0 15.6k
DavidAU/Qwen3-30B-A6B-16-Extreme finetune — 30.5B L — DavidAU 67 7.5k
Mungert/Qwen3-30B-A6B-16-Extreme-GGUF quantized GGUF — — — Mungert 2 8.9k
Mungert/Qwen3-30B-A3B-GGUF quantized GGUF — — apache-2.0 Mungert 4 8.6k
AlphaGaO/Qwen3-30B-A3B-GPTQ quantized GPTQ 30.5B L apache-2.0 AlphaGaO 3 8.2k
Aratako/Qwen3-30B-A3B-NSFW-JP finetune — 30.5B L mit Aratako 32 6k
DavidAU/Qwen3-30B-A1.5B-High-Speed finetune — 30.5B L — DavidAU 14 6.4k
Mungert/Pantheon-Proto-RP-1.8-30B-A3B-GGUF quantized GGUF — — apache-2.0 Mungert 1 7k
Gryphe/Pantheon-Proto-RP-1.8-30B-A3B finetune — 30.5B L apache-2.0 Gryphe 30 4.5k
Mungert/Qwen3-30B-A1.5B-High-Speed-GGUF quantized GGUF — — — Mungert 1 5.9k
mradermacher/Qwen3-30B-A3B-Base-i1-GGUF quantized GGUF — — apache-2.0 mradermacher 2 5.5k
NePe/Qwen3-30B-A3B-GPTQ quantized GPTQ 30.5B L apache-2.0 NePe 1 5.3k
mozilla-ai/Qwen3-30B-A3B-llamafile finetune — — — apache-2.0 mozilla-ai 7 4k
mradermacher/Qwen3-30B-A3B-Base-GGUF quantized GGUF — — apache-2.0 mradermacher 3 3.5k
Columbidae/Qwen3-16B-A3B-Base finetune — 16.0B L — Columbidae 1 3.4k
KaraKaraWarehouse/CavesOfQwen3 merge — 30.5B L apache-2.0 KaraKaraWarehouse 1 2.8k
tokyotech-llm/Qwen3-Swallow-30B-A3B-CPT-v0.2 finetune — 30.5B L apache-2.0 tokyotech-llm 0 2.8k
Userb1az/Qwen3-30B-A3B-GGUF quantized GGUF — — apache-2.0 Userb1az 0 2.4k
MegaScience/Qwen3-30B-A3B-MegaScience finetune — 30.5B L apache-2.0 MegaScience 10 1.8k
Azure99/Blossom-V6.3-30B-A3B-GGUF quantized GGUF — — apache-2.0 Azure99 0 2.2k
Antigma/Qwen3-30B-A3B-Base-GGUF quantized GGUF — — apache-2.0 Antigma 0 2.2k
DevQuasar/Qwen.Qwen3-30B-A3B-Base-GGUF quantized GGUF — — — DevQuasar 0 1.7k
brandonmusic/Qwen3-30B-A3B-ShapleyMCG-K34-Validation-Reconstruction finetune — 30.5B L apache-2.0 brandonmusic 0 1.2k
Qwen/SAE-Res-Qwen3-30B-A3B-Base-W128K-L0_100 finetune — — — other Qwen 8 819
hon9kon9ize/CantoneseLLM-v2.0-30B-A3B-Thinking-Chat-Vector-Merged merge — 30.5B L apache-2.0 hon9kon9ize 5 921
Qwen/SAE-Res-Qwen3-30B-A3B-Base-W32K-L0_50 finetune — — — other Qwen 7 776
hon9kon9ize/CantoneseLLM-v2.0-30B-A3B-Thinking finetune — 30.5B L apache-2.0 hon9kon9ize 3 969
Azure99/Blossom-V6.3-30B-A3B finetune — 30.5B L apache-2.0 Azure99 3 891
zjiayu064/Qwen3-30B-A3B-Base-BitsMoE-2bit quantized — 3.71B M apache-2.0 zjiayu064 0 988
ibnsina-llm/ibnsina-30b quantized GGUF 30.5B L apache-2.0 ibnsina-llm 0 763
zianglih/Qwen3-30B-A3B-MXFP8 quantized — 31.5B L apache-2.0 zianglih 0 694
YOYO-AI/Qwen3-30B-A3B-YOYO-V2 merge — 30.5B L apache-2.0 YOYO-AI 4 436
z369217411/Qwen3-30B-A3B-DAPO-BF16-step-0820 finetune -BF16 30.5B L — z369217411 0 601
YOYO-AI/Qwen3-30B-A3B-YOYO-V6 merge — 30.5B L apache-2.0 YOYO-AI 8 189
charlesthefool/Qwen3-30B-A3B-Base-Q4_K_M-GGUF quantized GGUF — — apache-2.0 charlesthefool 0 554
z369217411/Qwen3-30B-A3B-DAPO-BF16-step-0465 finetune -BF16 30.5B L — z369217411 0 554
z369217411/Qwen3-30B-A3B-DAPO-BF16-step-0620 finetune -BF16 30.5B L — z369217411 0 526
ForeverBlue/PrismQuant-Qwen3-30B-A3B-Base finetune — — — apache-2.0 ForeverBlue 1 460

Repository changelog

1 change
changed in the last 7 daysNew on the watch list: 2026-10-23all changes

Articles and coverage

10 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.

Analysis2

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.
All coverage tracked for this model (10) — append-only, nothing is ever removed
TitleSourcePublishedFirst seenSignals
Qwen3.5 122B and 35B models offer Sonnet 4.5 performance on local computersventurebeat.com2026-02-282026-10-14hn
Qwen3-Coder-Nextqwen.ai official2026-02-032026-10-14hn
Qwen3-Max-Thinkingqwen.ai official2026-01-262026-10-14hn
Qwen3-TTS family is now open sourced: Voice design, clone, and generationqwen.ai official2026-01-222026-10-14hn
Qwen3-VLqwen.ai official2025-09-232026-10-14hn
Qwen3-Omni: Native Omni AI model for text, image and videogithub.com2025-09-222026-10-14hn
Qwen3-Nextqwen.ai official2025-09-122026-10-14hn
GPT-OSS vs. Qwen3 and a detailed look how things evolved since GPT-2magazine.sebastianraschka.com2025-08-102026-10-14hn
Qwen3-Coder: Agentic coding in the worldqwenlm.github.io official2025-07-222026-10-14hn
Qwen3: Think deeper, act fasterqwenlm.github.io official2025-04-282026-10-14hn

Papers

1 paper

Papers 1

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).

Similar models

3 models

Similar models

  • GLM-4.7-Flash · 31.2B
    same size band (L), also MoE, same task (text-generation), more providers, permissive license
  • North-Mini-Code-1.0 · 30.5B
    same size band (L), also MoE, same task (text-generation), more providers, permissive license
  • Tongyi-DeepResearch-30B-A3B · 30.5B
    same size band (L), also MoE, same task (text-generation), permissive license

Sources

3 sources

Sources

Hub snapshot: 2026-10-25 · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:46 UTC · Summary: 2026-10-14 14:08 UTC · Last verified 2026-10-25

Follow Qwen3-30B-A3B-Base

Watch this model

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