[whatmodel.sh](../../README.md) / Qwen3 30B-A3B Thinking 2507
# Qwen3 30B-A3B Thinking 2507
Alibaba · released 2025-07 · 30.5B params (3.3B active) · apache-2.0
**score 60.2** ±0.5 · #198 of 732 · from 93 benchmark results across 14 sources
Score within the [LMArena (text)](https://whatmodel.sh/?set=lmarena-text&score=set) comparison set only.
## What it's good at
● marks Qwen3 30B-A3B Thinking 2507 among every model on that task or test, worst to best.
task vs the others rank counts tests
agent tasks ▁▁▃▆●▄▅▅▄▄▂▁ #186 of 313 21% 10
coding ▁▁▁▃▅▇█●▅▄▃▁ #195 of 469 4.4% 3
chat & writing ▁▁▁▃▆▄▆█●▇▇▃ #153 of 402 5.1% 2
math ▁▁▃▅█▆●█▄▂▁▁ #164 of 431 19% 10
agent tasks · #186 of 313 · 21% of its score
Beats [o1 (2024-12-17)](o1-2024-12-17.md), [o3-mini](o3-mini.md), [DeepSeek R1 0528](deepseek-r1-0528.md) here, though they rank higher overall. Trails [SmolLM3 3B](smollm3-3b.md), [Jamba 1.5 Mini](jamba-1-5-mini.md), [Jamba 1.5 Large](jamba-1-5-large.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
τ²-bench Telecom (maker) 26.3% ▃·●········█ beats 1/6 very hard · avg 9.8% 8.9% sharp · at its level · several runs
τ-bench Retail (maker) 67.8% ▁····▃·▂▇●▂▆ beats 18/28 medium · avg 37% 3.2% at its level · several runs
τ-bench Airline (maker) 48.0% ▁▁▁·▅▄▁█●▁▁▇ beats 18/28 hard · avg 25% 2.4% at its level · small run · several runs
τ²-bench Retail (maker) 58.8% ▁····▂▁●▆█▃▁ beats 10/44 medium · avg 37% 2.0% at its level · several runs
τ²-bench Airline (maker) 58.0% ▁▁·▂▁▃▂▁●▅▂▃ beats 24/42 hard · avg 32% 1.4% at its level · several runs
SWE-bench Verified 31.0% ▁▁▁▁▁▁●▄█▆▁▁ beats 19/138 hard · avg 16% 1.2% sharp · at its level · small run
BFCL v3 (maker) 72.4% ▁▁▁▁▁▃▅█▆▅●▁ beats 162/171 medium · avg 59% 0.7% blunt · at its level · several runs
BFCL V4 (December 2025 archive) (… 42.4% ▁▁▁▂▆▆▅●█▂▄▃ beats 55/114 medium · avg 35% 0.5% at its level
Multi-SWE-bench (scaffold per not… 9.5% █··●·······█ beats 1/2 very hard · avg 5.9% 0.5% small run
τ²-bench aggregate (maker) 41.9% ▁▁▁·●▁▃▃▃▄█▁ beats 3/39 medium · avg 40% 0.4% at its level · noisy
coding · #195 of 469 · 4.4% of its score
Beats [Gemini 2.5 Flash](gemini-2-5-flash.md) here, though they rank higher overall. Trails [QwQ-32B-Preview](qwq-32b-preview.md), [Ministral 3 8B Reasoning 2512](ministral-3-8b-reasoning.md), [OpenThinker 32B](openthinker-32b.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
LiveCodeBench v6 (Feb–May 2025) 60.3% ▁▁▃▁█▅●▇▅▅▃█ beats 17/38 medium · avg 41% 2.8% at its level · several runs
SWE-bench Verified 31.0% ▁▁▁▁▁▁●▄█▆▁▁ beats 19/138 hard · avg 16% 1.2% sharp · at its level · small run
Multi-SWE-bench (scaffold per not… 9.5% █··●·······█ beats 1/2 very hard · avg 5.9% 0.5% small run
chat & writing · #153 of 402 · 5.1% of its score
Beats [INTELLECT-3](intellect-3.md), [MiniMax-M2](minimax-m2.md), [Nemotron 3 Super 120B-A12B](nemotron-3-super-120b-a12b.md) here, though they rank higher overall. Trails [LFM2.5 1.2B Thinking](lfm2-5-1-2b-thinking.md), [LFM2.5 8B-A1B](lfm2-5-8b-a1b.md), [LFM2.5 2.6B](lfm2-5-2-6b.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
Arena-Hard v2 aggregate (maker) 56.0% ▁▁··▄▃▁●█▁▁▇ beats 8/22 hard · avg 28% 3.7% sharp · at its level · several runs
IFEval (aggregation unspecified) … 88.9% ▁▂▄▂▂▄▆█●▅▂▁ beats 104/145 easy · avg 78% 1.4% several runs
math · #164 of 431 · 19% of its score
Beats [MiMo-V2.5 (310B-A15B)](mimo-v2-5.md), [MiniMax-M2.7](minimax-m2-7.md), [MiniMax-M2.1](minimax-m2-1.md) here, though they rank higher overall. Trails [Olmo 3 7B Think](olmo-3-7b-think.md), [Qwen3-VL 4B Thinking](qwen3-vl-4b-thinking.md), [Nemotron 3 Nano 30B-A3B](nemotron-3-nano-30b-a3b.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
HMMT 2025 (edition unspecified) (… 71.4% ▃█▇●▇▅▁·▁··▁ beats 19/32 hard · avg 34% 4.3% sharp · at its level · several runs
AIME 2025 81.3% ▁·▁▁▁▂▅▆█●▅▁ beats 100/133 medium · avg 55% 3.4% sharp · small run · several runs
MathArena ArXivMath January 2026 … 39.1% ▂··●▂▄█▆▆▄▄█ beats 2/23 hard · avg 23% 1.9% at its level
MathArena ArXivMath February 2026… 22.7% ▁▁●▁█▇▄▃▁▃·▃ beats 2/24 very hard · avg 9.4% 1.9% typical
OTIS mock AIME 2024–2025 70.3% ▃▃▄▃▃▃●▇█▃▂▅ beats 101/197 very hard · avg 7.2% 1.8% sharp · at its level · noisy
PolyMATH (maker) 52.6% ▁···▃▁·▇●▄▄▄ beats 14/23 hard · avg 32% 1.6% at its level · several runs
AIME 2024 87.7% ▁▁▁▂▁▃▁▅██●▃ beats 65/85 easy · avg 69% 1.6% sharp
MathArena ArXivMath December 2025… 27.9% ●·▃▃█▁·▄·▁▁▁ beats 1/18 hard · avg 19% 1.1% at its level
HMMT February 2025 (maker) 63.1% ▁▁▁▃▁●▅▃██▁▄ beats 7/31 medium · avg 49% 1.1% at its level
HMMT November 2025 (maker) 73.8% ▁▁····●▅▃█▇▁ beats 2/23 medium · avg 62% 0.9% typical
other tests (35)
test result vs the others in its run difficulty counts why
GPQA Diamond 70.1% ▁▁▁▁▂▄▆▄●▇█▅ beats 105/211 hard · avg 39% 10% sharp · at its level · several runs
GPQA (subset unspecified) (maker) 73.4% ▂▁▂▂▄▃▆▃▅●▃▃ beats 85/106 hard · avg 49% 4.3% sharp · at its level · several runs
MMLU-Pro 81.9% ▁▁·▁▁▁▃▂▄▅●▁ beats 142/188 medium · avg 61% 3.6% several runs
MMLU-ProX (maker) 76.4% ▁·▁▁▁▂▁▃▃●▅▄ beats 16/36 medium · avg 61% 2.6% several runs
SuperGPQA (maker) 56.8% ▂▁▂▂▅▆▃▇●▇█▃ beats 30/46 medium · avg 39% 2.2% at its level · several runs
MMMLU (maker) 78.4% ▂▁▁▂▃▂▄▄▅●█▃ beats 29/46 medium · avg 63% 2.2% at its level · several runs
INCLUDE (maker) 74.4% ▂····▂▂▄▄●█▄ beats 21/36 medium · avg 59% 2.1% at its level · several runs
MMLU-Redux (maker) 91.4% ▁▁▃▁▂▃▄▅▇●█▄ beats 42/63 easy · avg 82% 2.0% several runs
Medmarks-V weighted mean win rate… 55.9% ▂▃▅▂▆▃█▅●▅▃▆ beats 25/34 medium · avg 41% 1.8% blunt · at its level
OJBench (maker) 25.1% ▁···▁▃·▅▄●██ beats 13/26 very hard · avg 13% 1.8% at its level · several runs
MMLU 86.9% ▁▁▁▁▁▂▂▅██▆● beats 198/214 medium · avg 70% 1.6% typical
BRIDGE clinical text, zero-shot a… 41.8% ▁▁▁▁▁▂▂▃▄▆●▃ beats 86/97 hard · avg 31% 1.4% blunt · at its level
KINA (Knowledge Index of Noah's A… 27.0% ▁▁▂▁▃▄●▆▄▅▂▃ beats 13/39 very hard · avg 21% 1.3% typical
C-Eval (maker) 87.4% ▁···▁▁▂▁▂▂●█ beats 18/33 medium · avg 73% 1.3% typical
LongBench v2 (maker) 44.8% ▂·▁·▂▃▁▄●▃█▄ beats 15/33 very hard · avg 35% 1.3% at its level
WritingBench (maker) 85.0% ▁·▁▂▁··▅●▂▂▁ beats 11/20 easy · avg 71% 1.3% several runs
Humanity's Last Exam 8.7% ▁▁▁▄●▆▇██▆▁▃ beats 18/111 very hard · avg 4.0% 1.1% typical
LiveBench 2024-11-25 (maker) 76.8% ▄··▂·▂·▂█●▂▂ beats 10/13 medium · avg 61% 1.1% at its level · several runs
Multi-IF (aggregation unspecified… 76.4% ▁▁▁▁▁▁▃▄●▄▁▂ beats 22/41 medium · avg 63% 1.0% several runs
Global PIQA (maker) 80.2% ▃▃··▃··▃●▅█▅ beats 4/11 medium · avg 71% 0.8% at its level
IFBench (aggregation unspecified)… 51.5% ▂▆▂▄▄▄●▄▄▆█▇ beats 37/86 medium · avg 39% 0.8% at its level
MultiChallenge (maker) 46.5% ▁·▁·▁▃▆●▃▂█▅ beats 12/30 hard · avg 31% 0.8% at its level
MAXIFE (maker) 77.4% ▂·▂·▂···●·▄█ beats 3/11 medium · avg 64% 0.8% typical
VitaBench (maker) 14.1% ▄··●▄▆▆█▄▄▄▂ beats 2/21 very hard · avg 11% 0.7% typical
Creative Writing v3 (maker) 84.4% ▂····▂▆··●█▆ beats 6/14 easy · avg 71% 0.7% small run
KOR-Bench 70.2% ▄·······●··▄ beats 1/3 medium · avg 55% 0.6% at its level
SimpleQA 23.6% ▄▄▄▆█▆█●█·▂█ beats 21/32 very hard · avg 6.1% 0.6% typical
NOVA-63 (maker) 52.5% ▃▃·▃···●█▃▃█ beats 3/11 medium · avg 39% 0.5% blunt · at its level
Chess Puzzles (Epoch) 8.0% █▁▁●▁▂▃▃▃▂▁▁ beats 54/137 very hard · avg 5.8% 0.4% sharp · too hard for it · noisy
RULER 128k 94.5% ▃··▅▄█▅█▇·▃● beats 32/33 easy · avg 72% 0.4% typical
BeyondAIME 56.0% ▄·▂·▂▄··●▄▄█ beats 6/15 medium · avg 37% 0.2% at its level · noisy
DeepPlanning (maker) 4.9% ▂·····●··▄█▄ beats 1/9 very hard · avg 7.1% 0.2% noisy
Vectara hallucination leaderboard… 95.2% ▁▁▂▃▄▄●▆▃▂▂▁ beats 64/140 very easy · avg 94% 0.1% blunt · maxed out · noisy
WMT24++ (maker) 69.3% ▁·▁▁·▁·▁●▆█▃ beats 5/27 easy · avg 66% 0.1% blunt · at its level
miniF2F (pass@1) 5.7% ●·█▅·····▅·▃ beats 0/8 very hard · avg 14% 0.1% blunt · noisy · small run
source results share of the score
Qwen/Qwen3-Next-80B-A3B-Thinking model … 19 23% ▇▇▇
Qwen/Qwen3-30B-A3B-Thinking-2507 model … 23 21% ▇▇▇
ByteDance-Seed/Seed-OSS-36B-Instruct mo… 18 18% ▇▇
Qwen/Qwen3.5-9B model card (submitted /… 16 12% ▇
Epoch AI 3 6% ▇
Qwen/Qwen3.5-4B model card (submitted /… 3 6% ▇
MathArena ArXivMath / BrokenArXiv / ArX… 3 5% ▇
Qwen (submitted / maker-reported) 1 4% ▇
Medmarks: A Comprehensive Open-Source L… 1 2% ▇
BRIDGE Medical Leaderboard 1 1% ▇
Knowledge Index of Noah's Ark 1 1% ▇
NVIDIA 2 1% ▇
LiquidAI/LFM2.5-8B-A1B model card 1 0% ▇
Vectara hallucination leaderboard 1 0% ▇
prior: pull toward the middle 0% ▇
weighted average = 60.2
Licences: Qwen/Qwen3-Next-80B-A3B-Thinking model card ([apache-2.0](../../licenses/Qwen--Qwen3-Next-80B-A3B-Thinking.txt)) · Qwen/Qwen3-30B-A3B-Thinking-2507 model card ([apache-2.0](../../licenses/Qwen--Qwen3-30B-A3B-Thinking-2507.txt)) · ByteDance-Seed/Seed-OSS-36B-Instruct model card ([apache-2.0](../../licenses/ByteDance-Seed--Seed-OSS-36B-Instruct.txt)) · Qwen/Qwen3.5-9B model card ([apache-2.0](../../licenses/Qwen--Qwen3.5-9B.txt)) · Epoch AI ([CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)) · Qwen/Qwen3.5-4B model card ([apache-2.0](../../licenses/Qwen--Qwen3.5-4B.txt)) · MathArena ArXivMath / BrokenArXiv / ArXivLean model-output datasets ([CC BY-SA 4.0](../../licenses/matharena-sharealike-NOTICE.txt)) · Medmarks: A Comprehensive Open-Source LLM Benchmark Suite for Medical Tasks ([CC BY 4.0](../../licenses/paper-2605.01417v1-CC-BY-4.0.txt)) · BRIDGE Medical Leaderboard ([MIT](../../licenses/bridge-medical-leaderboard-MIT.txt)) · Knowledge Index of Noah's Ark ([CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)) · NVIDIA ([CC BY 4.0](../../licenses/report-2512.20848v1.txt)) · LiquidAI/LFM2.5-8B-A1B model card ([LFM Open License v1.0](../../licenses/cards-gap-NOTICE.txt)) · Vectara hallucination leaderboard ([Apache-2.0](https://www.apache.org/licenses/LICENSE-2.0))
## How to run it
### Through an API
api (Alibaba Cloud Model Studio) · $0.75/M blended · $0.2 in / $2.4 out per 1M tokens · [alibabacloud.com](https://www.alibabacloud.com/help/en/model-studio/model-pricing)
[price source](https://www.alibabacloud.com/help/en/model-studio/model-pricing): USD per 1M tokens, Alibaba Cloud Model Studio, Singapore (International deployment scope), qwen3-30b-a3b-thinking-2507 (thinking mode only; output = chain of thought + answer). Chinese-mainland/Global deployments priced differently; free quota, batch and context-cache discounts excluded. Currency: USD. Alibaba Cloud official Model Studio pricing page, observed 2026-09-25. · [context source](https://huggingface.co/Qwen/Qwen3-30B-A3B-Thinking-2507/blob/144afc2f379b542fdd4e85a1fcd5e1f79112d95d/README.md): Native context for this exact July 2025 model version.
### llama.cpp (recommended)
1. Install llama.cpp. Homebrew and winget installs name the commands llama-cli and llama-server instead. [Other ways to install](https://github.com/ggml-org/llama.cpp/blob/master/docs/install.md)
```sh
# macOS / Linux
curl -LsSf https://llama.app/install.sh | sh
# Windows
irm https://llama.app/install.ps1 | iex
```
2. Run Qwen3 30B-A3B Thinking 2507 (Q4_K_M, needs ~18.6 GB, from [huggingface.co/lmstudio-community](https://huggingface.co/lmstudio-community/Qwen3-30B-A3B-Thinking-2507-GGUF/tree/3397dd3a175b1767204c0d1a2d775ab0869cbab3))
```sh
llama cli -hf lmstudio-community/Qwen3-30B-A3B-Thinking-2507-GGUF:Q4_K_M
```
3. Or start a local server and open the address it prints (http://localhost:9931)
```sh
llama serve -hf lmstudio-community/Qwen3-30B-A3B-Thinking-2507-GGUF:Q4_K_M
```
Other versions: Q8_0 (~32.5 GB) `llama cli -hf lmstudio-community/Qwen3-30B-A3B-Thinking-2507-GGUF:Q8_0`
### LM Studio
1. Install LM Studio. Open the app once before using the lms command. [Download it (macOS / Windows / Linux)](https://lmstudio.ai/download)
2. Run Qwen3 30B-A3B Thinking 2507 (Q4_K_M, needs ~18.6 GB, from [huggingface.co/lmstudio-community](https://huggingface.co/lmstudio-community/Qwen3-30B-A3B-Thinking-2507-GGUF/tree/3397dd3a175b1767204c0d1a2d775ab0869cbab3))
```sh
lms get https://huggingface.co/lmstudio-community/Qwen3-30B-A3B-Thinking-2507-GGUF@Q4_K_M -y
```
3. Then run lms ls to see its name, and lms chat followed by that name
Other versions: Q8_0 (~32.5 GB) `lms get https://huggingface.co/lmstudio-community/Qwen3-30B-A3B-Thinking-2507-GGUF@Q8_0 -y` · 4bit (~17.2 GB, Mac only) `lms get https://huggingface.co/lmstudio-community/Qwen3-30B-A3B-Thinking-2507-MLX-4bit -y` · 8bit (~32.4 GB, Mac only) `lms get https://huggingface.co/lmstudio-community/Qwen3-30B-A3B-Thinking-2507-MLX-8bit -y`
### Ollama
1. Install Ollama. [Download it (macOS / Windows)](https://ollama.com/download)
```sh
# Linux
curl -fsSL https://ollama.com/install.sh | sh
```
2. Run Qwen3 30B-A3B Thinking 2507 (Q4_K_M, needs ~18.8 GB, from [ollama.com](https://ollama.com/library/qwen3/tags))
```sh
ollama run qwen3:30b
```
Other versions: Q8_0 (~37.1 GB) `ollama run qwen3:30b-a3b-thinking-2507-q8_0`
### MLX (Mac)
1. Install MLX (Mac). Apple silicon Macs only, text only.
```sh
# with uv
uv tool install mlx-lm
# or
pip install mlx-lm
```
2. Run Qwen3 30B-A3B Thinking 2507 (4bit, needs ~17.2 GB, Mac only, from [huggingface.co/lmstudio-community](https://huggingface.co/lmstudio-community/Qwen3-30B-A3B-Thinking-2507-MLX-4bit/tree/02b2aef74a0f3168c9a189e98aaf9e240c2f5a72))
```sh
mlx_lm.chat --model lmstudio-community/Qwen3-30B-A3B-Thinking-2507-MLX-4bit
```
3. Or start an OpenAI-compatible server at http://localhost:8080/v1
```sh
mlx_lm.server --model lmstudio-community/Qwen3-30B-A3B-Thinking-2507-MLX-4bit
```
Other versions: 8bit (~32.4 GB, Mac only) `mlx_lm.chat --model lmstudio-community/Qwen3-30B-A3B-Thinking-2507-MLX-8bit`
Model repository and documentation: [huggingface.co/Qwen](https://huggingface.co/Qwen/Qwen3-30B-A3B-Thinking-2507)
## Quantization evidence
How much of the full-precision result a quantized version kept, measured within one study. Rows from the same study share a table and aren't independent. They don't describe every download of that precision.
precision test full → quantized kept
INT4 W4A16 (weight-only INT4,… MATH-500 — published row 'Math 50… 0.907 → 0.931 102.6%
INT4 W4A16 (weight-only INT4,… IFEval — published row 'IfEval (0… 0.901 → 0.903 100.2%
INT4 W4A16 (weight-only INT4,… GPQA Diamond — published row 'GPQ… 0.715 → 0.716 100.1%
INT4 W4A16 (weight-only INT4,… MMLU-Pro — published row 'MMLU-Pr… 0.807 → 0.791 98.0%
INT4 W4A16 (weight-only INT4,… AIME 2025 — published row 'AIME 2… 0.867 → 0.833 96.1%
INT4 W4A16 (weight-only INT4,… GSM8K-Platinum — published row 'G… 0.856 → 0.727 84.9%
INT8 W8A8 (weights INT8 per-c… GPQA Diamond — published row 'GPQ… 0.715 → 0.724 101.3%
INT8 W8A8 (weights INT8 per-c… IFEval — published row 'IfEval (0… 0.901 → 0.901 100.0%
INT8 W8A8 (weights INT8 per-c… MMLU-Pro — published row 'MMLU-Pr… 0.807 → 0.807 99.9%
INT8 W8A8 (weights INT8 per-c… MATH-500 — published row 'Math 50… 0.907 → 0.906 99.9%
INT8 W8A8 (weights INT8 per-c… GSM8K-Platinum — published row 'G… 0.856 → 0.838 98.0%
INT8 W8A8 (weights INT8 per-c… AIME 2025 — published row 'AIME 2… 0.867 → 0.842 97.1%
Each kept value links to its source.
[interactive version](https://whatmodel.sh/model/qwen3-30b-a3b-thinking-2507) · [JSON](https://whatmodel.sh/api?format=json&all=1)