whatmodel.sh / Qwen3 30B-A3B (Apr 2025, hybrid)

Qwen3 30B-A3B (Apr 2025, hybrid)

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Qwen3 30B-A3B (Apr 2025, hybrid) is an open-weights language model from Alibaba, released April 2025. Its published size is 30.5B parameters (3.3B active per token) and a 33k-token context window. Licence: apache-2.0.

Catalogue identity notes

Comparison model from the Epoch AI internal-evaluation CSVs (Epoch release date 2025-04-29). Original April 2025 hybrid-thinking checkpoint Qwen/Qwen3-30B-A3B; distinct from the 2507 Instruct/Thinking checkpoints. Default (thinking) runs are used; '_none' runs only where no default run exists. Card: 30.5B total, 3.3B activated. Beware: ollama qwen3:30b and qwen3:30b-a3b now point to the Thinking-2507 build, so the original tags are pinned explicitly (40K context).

55.2±0.5#261 of 719 LLMsfrom 112 benchmark results across 16 sources

$ how it compares

The LLMs ranked either side of Qwen3 30B-A3B (Apr 2025, hybrid), out of 828 with a score, then more from Alibaba.

#modelscorewhereRAMreleased
311Ministral 3 8B Reasoning 2512 Mistral55±1local~6.5GB2025-12
312Magistral Small 1.0 (2506) Mistral55±1local~16GB2025-06
313Qwen3 30B-A3B (Apr 2025, hybrid) Alibaba55±1api $0.35/M +local~19GB2025-04
314Molmo 2 8B Ai255±2local~6.8GB2025-12
315Magistral Medium 1.1 (2507) Mistral55±2comparison only2025-07
more from Alibaba
18Qwen3.8 Max (2.4T-A95B) Alibaba78±1api $3.00/M +local~380GB2026-08
34Qwen3.8 Max (0902) Alibaba76±1api $3.00/M2026-09
37Qwen3.8-Flash-Next Alibaba75±1local~59GB2026-08

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #205 of 300 · 22% of its score · 6 tests

Beats o1 (2024-12-17), o3-mini, DeepSeek R1 (Jan '25) here, though they rank higher overall. Trails SmolLM3 3B, Jamba 1.5 Mini, Mellum2 12B-A2.5B Instruct here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
τ²-bench Telecom lab +422.8% ✎▃·█········█ beats 0/6very hard · avg 14%▇▇▇▇ 9.1%sharp · small run · several runs
τ-bench Retail lab +338.3% ✎▁····▃·▂▇█▂▆ beats 12/28medium · avg 36%▇▇ 3.9%at its level · several runs
τ²-bench Retail lab +331.6% ✎▁····▃▁▃▆█▃▁ beats 1/44hard · avg 30%▇ 2.8%at its level · small run · several runs
τ-bench Airline lab +318.0% ✎▁▁▁·▅▄▁█▇▁▁▇ beats 4/28hard · avg 28%▇ 2.5%at its level · small run · several runs
τ²-bench Airline lab +318.0% ✎▁▁·▃▁▃▂▁█▅▂▃ beats 2/42hard · avg 18%▇ 2.1%at its level · small run · several runs
BFCL v3 +465.7% ✎▁▁▁▁▁▃▅█▅▄▁▁ beats 99/114medium · avg 52%▇ 1.5%blunt · at its level · several runs
coding▁▁▁▃▆██▇▆▄▃▁ #240 of 458 · 10% of its score · 3 tests

Beats Qwen3-VL 32B Instruct, Solar Open 100B (102B-A12B), GLM-4.7-Flash here, though they rank higher overall. Trails QwQ-32B-Preview, InternLM2.5 20B Chat, Mistral Small Instruct 22B (2409) here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
Aider polyglot (225 exercises, two attempts) lab +124.4% ✎▂▁▁▃▆▃▄▆█▃▄▂ beats 24/71very hard · avg 8.9%▇▇ 5.1%sharp · small run · several runs
LiveCodeBench v6 (Feb–May 2025) lab +357.4% ✎▁▁▃▁█▅▄▇▅▅▃█ beats 12/38hard · avg 28%▇ 3.2%at its level · small run · several runs
MultiPL-E lab +174.6% ✎▃··▄·▅▅▇▄█▃▄ beats 21/31easy · avg 66%▇ 1.3%at its level · small run · several runs
chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #189 of 393 · 11% of its score · 5 tests

Beats Seed-OSS 36B Instruct, Granite 4.2 8B, Qwen3-VL 30B-A3B Thinking here, though they rank higher overall. Trails LFM2.5 1.2B Instruct, LFM2.5 1.2B Thinking, Gemma 3 12B here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
Arena-Hard v2 aggregate lab +336.3% ✎▁▁··▄▃▁▃█▁▁▇ beats 4/22hard · avg 27%▇▇ 4.9%sharp · at its level · small run · several runs
Arena-Hard v2.0 hard prompts (style control, Gemini-2.5 judge)33.9% ✎▂·▃▅▅·▃█▆▂▂▂ beats 10/22very hard · avg 12%▇ 2.2%sharp
LMArena (text)1326.5 ✎▁▁▁▂▃▃▃▆▆▇█▃ beats 164/379avg 1268▇ 1.9%blunt
IFEval (aggregation unspecified) lab +386.5% ✎▁▁▃▁▂▄▄▆█▃▂▁ beats 83/141easy · avg 77%▇ 1.9%small run · several runs
Arena-Hard v2.0 creative writing (GPT-4.1 + Gemini-2.5 judges)34.9% ✎▆▂▂▆▃▂▃█▂▃·▂ beats 10/22hard · avg 19%▇ 0.5%sharp · at its level · noisy
math▁▂▃▅▆▅▄█▄▁▁▁ #191 of 398 · 15% of its score · 4 tests

Beats MiniMax-M2.1, Qwen3 235B-A22B (Apr 2025, hybrid), DeepSeek V3 (Dec '24) here, though they rank higher overall. Trails Phi-4-mini-reasoning 3.8B, DeepSeek R1 Distill Qwen 7B, OpenThinker3 7B here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
HMMT 2025 (edition unspecified) lab +349.8% ✎▃█▇▅▇▅▁·▁··▁ beats 8/32hard · avg 33%▇▇ 6.0%sharp · at its level · small run · several runs
AIME 2025 lab +370.9% ✎▁·▁▁▁▂▅▆█▇▅▁ beats 66/133medium · avg 55%▇▇ 5.6%sharp · at its level · small run · several runs
PolyMATH lab +346.1% ✎▁···▃▁·▇█▄▄▄ beats 7/23hard · avg 34%▇ 2.5%at its level · small run · several runs
OTIS mock AIME 2024–202562.8% ✎▃▃▄▃▃▃▇▇█▃▂▅ beats 91/193very hard · avg 8.5%▇ 1.3%sharp · noisy
other tests22 not in a task · 42% of its score
testresultvs the others in its rundifficultycountswhy
GPQA (subset unspecified) lab +365.8% ✎▂▁▂▂▄▃▆▃▅█▄▃ beats 69/104medium · avg 51%▇▇▇ 6.5%sharp · at its level · small run · several runs
MMLU-Pro lab +378.5% ✎▁▁·▁▁▁▃▂▄▅█▁ beats 119/188medium · avg 59%▇▇ 4.0%at its level · small run · several runs
MMLU-Redux lab +389.5% ✎▁▁▃▁▂▃▄▅▇▅█▄ beats 34/63easy · avg 79%▇ 3.6%small run · several runs
INCLUDE lab +371.9% ✎▂····▂▂▄▄▄█▄ beats 19/36medium · avg 58%▇ 3.6%at its level · small run · several runs
GPQA Diamond61.7% ✎▁▁▁▁▂▄▆▄▅▇█▅ beats 89/208hard · avg 40%▇ 3.1%sharp · at its level
SuperGPQA lab +351.8% ✎▂▁▂▂▅▆▃▇▇▇█▃ beats 21/46hard · avg 35%▇ 2.4%at its level · small run · several runs
Creative Writing v3 lab +379.1% ✎▂····▂▆··▆█▆ beats 4/14easy · avg 70%▇ 2.4%small run · several runs
MMLU-ProX lab +373.1% ✎▁·▁▁▁▂▁▃▂█▅▄ beats 11/35medium · avg 57%▇ 2.3%at its level · small run · several runs
14 more tests
testresultvs the others in its rundifficultycountswhy
LiveBench 2024-11-25 lab +374.3% ✎▄··▂·▂·▃█▆▂▂ beats 4/13medium · avg 62%▇ 2.0%at its level · small run · several runs
WritingBench lab +377.0% ✎▁·▃▂▁··▅█▂▂▁ beats 1/20easy · avg 70%▇ 2.0%small run · several runs
Multi-IF (aggregation unspecified) lab +372.2% ✎▁▁▁▁▁▁▃▄█▄▁▂ beats 15/41medium · avg 63%▇ 1.8%blunt · at its level · small run · several runs
OJBench lab +120.7% ✎▁···▁▃·▅▄▅██ beats 9/26very hard · avg 13%▇ 1.5%small run · several runs
EQ-Bench 3 (emotional intelligence, judge Elo)1082.9 ✎▁·▂▁▄▅▅▃█▄▁▁ beats 10/62avg 1044▇ 1.2%typical
CPTUB average (SpeakLeash, 0-5 judge scale)70.7% ✎▁▂▂▂▃█▃▆▅▄▂▃ beats 55/95easy · avg 66%▇ 1.1%at its level
BRIDGE clinical text, zero-shot average (87 tasks)40.9% ✎▁▁▁▁▁▂▂▃▄▆█▃ beats 80/97hard · avg 32%▇ 1.0%blunt · at its level
KINA (Knowledge Index of Noah's Ark) accuracy, 899 ten-option items24.3% ✎▁▁▂▁▃▄█▆▄▅▂▃ beats 8/39very hard · avg 22%▇ 0.9%blunt
Open Persian LLM Leaderboard v2 average (21 tasks)55.2% ✎▂▄▂▃█▄▃▆▃▅▂▂ beats 42/81medium · avg 58%▇ 0.8%blunt · at its level
ZebraLogic lab +133.2% ✎█·▁▁▁▇·▄▄▁▄▁ beats 2/24medium · avg 49%▇ 0.6%at its level · noisy · small run · several runs
Open PT-LLM Leaderboard average (9 Portuguese tasks)76.8% ✎▃▁▁▁▁▁▁▁▂█▅▁ beats 237/276medium · avg 73%▇ 0.4%at its level · noisy
EuroEval HellaSwag (English), test split, 5-shot generative MC, accuracy81.7% ✎▃·▁▁▂▁▂▅█▇▄▁ beats 76/100medium · avg 68%▇ 0.3%at its level · noisy
EuroEval Life in the UK (English knowledge MC), test split, 5-shot, accuracy94.8% ✎▁▁▁▁▁▁▁▂▄█▄▃ beats 140/146easy · avg 83%▇ 0.2%noisy
Chess Puzzles (Epoch)4.0% ✎█▁▁▁▁▂▁▄▂▂▁▁ beats 3/135very hard · avg 5.9%▇ 0.2%sharp · too hard for it · noisy

$ where to run it

Through an API

api · Alibaba Cloud Model Studio · $0.35/M · $0.2 in / $0.8 out per 1M tokens · 33k context · www.alibabacloud.com ↗ · price sourceUSD per 1M tokens, Singapore (International), qwen3-30b-a3b (hybrid; non-thinking output price; thinking-mode output is $2.4), single tier: input $0.2, output $0.8. Batch and context-cache discounts, higher tiers and free quota excluded. Observed 2026-09-28.

llama.cpp · recommended

version

  1. 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 ↗

  2. Run Qwen3 30B-A3B (Apr 2025, hybrid) · needs ~18.6 GB · from huggingface.co/Qwen (official)
    chat in the terminal
    llama cli -hf Qwen/Qwen3-30B-A3B-GGUF:Q4_K_M
  3. Or start a local server and open the address it prints (http://localhost:9931)
    llama serve -hf Qwen/Qwen3-30B-A3B-GGUF:Q4_K_M

LM Studio

version

  1. Install LM Studio · Open the app once before using the lms command.

    download it (macOS / Windows / Linux) ↗

  2. Run Qwen3 30B-A3B (Apr 2025, hybrid) · needs ~18.6 GB · from huggingface.co/Qwen (official)

    open in LM Studio ↗

  3. Or download it in the terminal
    lms get https://huggingface.co/Qwen/Qwen3-30B-A3B-GGUF@Q4_K_M -y
  4. Then run lms ls to see its name, and lms chat followed by that name

Ollama

version

  1. Install Ollama

    download it (macOS / Windows) ↗

    Linux
    curl -fsSL https://ollama.com/install.sh | sh
  2. Run Qwen3 30B-A3B (Apr 2025, hybrid) · needs ~18.8 GB · from ollama.com
    chat in the terminal
    ollama run qwen3:30b-a3b-q4_K_M

MLX (Mac)

version

  1. Install MLX (Mac) · Apple silicon Macs only, text only.
    with uv
    uv tool install mlx-lm
    or
    pip install mlx-lm
  2. Run Qwen3 30B-A3B (Apr 2025, hybrid) · needs ~16.2 GB, Mac only · from huggingface.co/Qwen (official)
    chat in the terminal
    mlx_lm.chat --model Qwen/Qwen3-30B-A3B-MLX-4bit
  3. Or start an OpenAI-compatible server at http://localhost:8080/v1
    mlx_lm.server --model Qwen/Qwen3-30B-A3B-MLX-4bit

Official weights and documentation: huggingface.co/Qwen ↗

$ how the score was worked out · Qwen/Qwen3-4B-Instruct-2507 m… 21%, Qwen/Qwen3-30B-A3B-Thinking-2… 21%, Qwen/Qwen3-4B-Thinking-2507 m… 21% and 13 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

sourceresultsshare of the scorelicence
Qwen/Qwen3-4B-Instruct-2507 model card · lab-reported25▇▇▇▇▇▇ 21%apache-2.0
Qwen/Qwen3-30B-A3B-Thinking-2507 model card · lab-reported23▇▇▇▇▇▇ 21%apache-2.0
Qwen/Qwen3-4B-Thinking-2507 model card · lab-reported23▇▇▇▇▇▇ 21%apache-2.0
Qwen/Qwen3-30B-A3B-Instruct-2507 model card · lab-reported25▇▇▇▇▇▇ 19%apache-2.0
Epoch AI — internal evaluations3▇ 5%CC BY 4.0
Arena-Hard-Auto v2.0-Preview leaderboard (repository README)2▇ 3%Apache-2.0
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-reported1▇ 2%Factual vendor-reported score (not licensed content)
LMArena leaderboard dataset1▇ 2%CC BY 4.0
EQ-Bench 3 canonical leaderboard results1▇ 1%MIT
SpeakLeash CPTUB (Complex Polish Text Understanding Benchmark)1▇ 1%MIT
BRIDGE Medical Leaderboard (Harvard Medical School / Brigham and Women's YLab)1▇ 1%MIT
Knowledge Index of Noah's Ark (arXiv:2606.05104v2)1▇ 1%CC BY 4.0
Open Persian LLM Leaderboard v2 (Part DP AI and AUT NLP Lab)1▇ 1%Apache-2.0
EuroEval leaderboards (results collection), English generative leaderboard2▇ 0%MIT
Berkeley Function Calling Leaderboard result archive1▇ 0%Apache-2.0
Open Portuguese LLM Leaderboard (archived)1▇ 0%Apache-2.0
prior · pull toward the middle▇ 0%
weighted average= 55.2

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated · sources and licences · json