[whatmodel.sh](../../README.md) / Qwen3 235B-A22B (Apr 2025, hybrid)
# Qwen3 235B-A22B (Apr 2025, hybrid)
Alibaba · released 2025-04 · 235B params (22B active) · apache-2.0
**score 59.6** ±0.4 · #209 of 732 · from 104 benchmark results across 20 sources
Score within the [LMArena (text)](https://whatmodel.sh/?set=lmarena-text&score=set) comparison set only.
## What it's good at
● marks Qwen3 235B-A22B (Apr 2025, hybrid) among every model on that task or test, worst to best.
task vs the others rank counts tests
agent tasks ▁▁▃▆●▄▅▅▄▄▂▁ #200 of 313 25% 6
coding ▁▁▁▃▅▇█●▅▄▃▁ #163 of 469 14% 5
chat & writing ▁▁▁▃▆▄▆█●▇▇▃ #139 of 402 10% 5
math ▁▁▃▅█●▄█▄▂▁▁ #209 of 431 15% 5
agent tasks · #200 of 313 · 25% 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), [Mellum2 12B-A2.5B Instruct](mellum2-12b-a2-5b-instruct.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
τ²-bench Telecom 22.1% █●··█·██··██ beats 1/6 very hard · avg 7.7% 11% sharp · at its level · several runs
τ²-bench Retail 51.4% ▁····▂▁▃●█▃▁ beats 18/44 medium · avg 37% 4.1% at its level · several runs
τ-bench Retail (maker) 65.2% ▁····▃·▂▇●▂▆ beats 14/28 medium · avg 40% 4.0% at its level · small run · several runs
τ²-bench Airline 36.8% ▁▁·●▁▃▂▁█▅▂▃ beats 3/42 hard · avg 29% 3.2% at its level · several runs
τ-bench Airline (maker) 32.0% ▁▁▁·●▄▁█▇▁▁▇ beats 6/28 hard · avg 21% 2.5% at its level · several runs
BFCL v3 67.9% ▁▁▁▁▁▃▅█▅●▁▁ beats 104/114 medium · avg 51% 1.3% blunt · at its level · several runs
coding · #163 of 469 · 14% of its score
Beats [Qwen3-Coder-Next (80B-A3B)](qwen3-coder-next.md), [Gemini 2.5 Pro](gemini-2-5-pro.md), [Gemini 3.1 Flash-Lite](gemini-3-1-flash-lite.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), [Phi-4-reasoning 14B](phi-4-reasoning.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
Aider polyglot (225 exercises, tw… 59.6% ▂·▂▃▅▂▃█●▄▂▂ beats 34/49 very hard · avg 7.6% 7.6% sharp · at its level · several runs
LiveCodeBench v6 (Feb–May 2025) (… 55.7% ▁▁▃▁●▅▄▇▅▅▃█ beats 9/38 hard · avg 25% 2.7% at its level · small run · several runs
LiveCodeBench official (Aug 1, 20… 65.9% ▃▃▅▃··▃▅●▅▅▃ beats 9/15 hard · avg 24% 1.4% sharp · at its level
WebApp1K-Duo-React pass@1 (onekq-… 32.9% ▁··▁·●▁▁▆█▇█ beats 3/45 hard · avg 29% 1.3% at its level
MultiPL-E (maker) 79.3% ▃··▄·▅▅▇▄●▃▄ beats 25/31 easy · avg 66% 0.6% small run
chat & writing · #139 of 402 · 10% of its score
Beats [o3-mini](o3-mini.md), [INTELLECT-3](intellect-3.md), [MiniMax-M2](minimax-m2.md) here, though they rank higher overall. Trails [LFM2.5 8B-A1B](lfm2-5-8b-a1b.md), [LFM2.5 2.6B](lfm2-5-2-6b.md), [Nemotron-Cascade-8B](nemotron-cascade-8b.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
Arena-Hard v2 aggregate (maker) 61.5% ▁▁··▄▃▁▃●▁▁▇ beats 10/22 hard · avg 27% 3.7% sharp · at its level · small run · seve…
Arena-Hard v2.0 hard prompts (sty… 58.4% ▂·▃▅▅·▃█●▂▂▂ beats 17/22 very hard · avg 12% 2.4% sharp · at its level
LMArena (text) 1374.9 ▁▁▁▂▃▃▃▆●▇█▃ beats 217/385 avg 1260 2.1% blunt
IFEval (aggregation unspecified) … 83.4% ▁▂▄▂▂▄●██▅▂▁ beats 66/145 easy · avg 77% 1.4% several runs
Arena-Hard v2.0 creative writing … 73.5% ▆▂▂▆▂▂▃█▂●·▂ beats 20/22 hard · avg 19% 0.5% sharp · at its level · noisy
math · #209 of 431 · 15% of its score
Beats [MiniMax-M2.1](minimax-m2-1.md), [DeepSeek V3 (Dec '24)](deepseek-v3.md) here, though they rank higher overall. Trails [Granite 8B Code Base (4K)](granite-8b-code-base-4k.md), [Mixtral 8x7B v0.1 (base)](mixtral-8x7b-v0-1.md), [Olmo 3 7B Instruct](olmo-3-7b-instruct.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
AIME 2025 (maker) 81.5% ▁·▁▁▁▂▅▆█●▅▁ beats 88/133 medium · avg 55% 4.8% sharp · small run · several runs
HMMT 2025 (edition unspecified) (… 62.5% ▃█●▅▇▅▁·▁··▁ beats 15/32 hard · avg 33% 4.4% sharp · at its level · small run · seve…
MATH level 5 68.9% ▂▂▅▆▄█●▃▅▂▃▄ beats 59/92 hard · avg 28% 3.5% sharp
PolyMATH (maker) 54.7% ▁···▃▁·▇█●▄▄ beats 15/23 hard · avg 34% 1.7% at its level · small run · several runs
THINK-Bench final-answer accuracy… 94.9% ▄█·▄····▄▄·● beats 7/7 easy · avg 76% 0.3% typical
other tests (23)
test result vs the others in its run difficulty counts why
GPQA (subset unspecified) (maker) 71.1% ▂▁▂▂▄▃▆▃▅●▃▃ beats 80/106 hard · avg 50% 4.7% sharp · at its level · small run · seve…
MMLU-Pro (maker) 82.8% ▁▁·▁▁▁▃▂▄▅●▁ beats 153/188 medium · avg 58% 3.9% at its level · several runs
GPQA Diamond 70.7% ▁▁▁▁▂▄▆▄●▇█▅ beats 106/211 hard · avg 39% 3.5% sharp · at its level
INCLUDE (maker) 78.7% ▂····▂▂▄▄▄●▄ beats 23/36 medium · avg 59% 2.9% at its level · several runs
MMLU-Redux (maker) 92.7% ▁▁▃▁▂▃▄▅▇▅●▄ beats 48/63 easy · avg 81% 2.8% several runs
MMLU-ProX (maker) 80.0% ▁·▁▁▁▂▁▃▃●▅▄ beats 24/36 medium · avg 60% 1.8% at its level · several runs
Creative Writing v3 (maker) 84.6% ▂····▂▆··●█▆ beats 7/14 easy · avg 71% 1.8% small run · several runs
SuperGPQA (maker) 60.7% ▂▁▂▂▅▆▃▇▇●█▃ beats 35/46 hard · avg 35% 1.7% at its level · small run · several runs
WritingBench (maker) 80.3% ▁·▁▂●··▅█▂▂▁ beats 4/20 easy · avg 66% 1.5% several runs
Multi-IF (aggregation unspecified… 71.9% ▁▁▁▁▁▁▃●█▄▁▂ beats 14/41 medium · avg 60% 1.4% at its level · several runs
OJBench (maker) 25.6% ▁···▁▃·▅▄●██ beats 14/26 very hard · avg 13% 1.4% at its level · small run · several runs
LiveBench 2024-11-25 (maker) 77.1% ▄··▂·▂·▂█●▂▂ beats 11/13 medium · avg 63% 1.3% small run · several runs
BRIDGE clinical text, zero-shot a… 41.6% ▁▁▁▁▁▂▂▃▄▆●▃ beats 84/97 hard · avg 31% 1.3% blunt · at its level
EQ-Bench 3 (emotional intelligenc… 1215.9 ▁·▂▁▄▅▅●█▄▁▁ beats 32/62 avg 1034 1.3% typical
KINA (Knowledge Index of Noah's A… 29.4% ▁▁▂▁▃▄●▆▄▅▂▃ beats 18/39 very hard · avg 21% 1.2% typical
CPTUB average (SpeakLeash, 0-5 ju… 78.2% ▁▂▂▂▃█▃▆●▄▂▃ beats 76/95 medium · avg 65% 1.2% typical
Humanity's Last Exam (maker) 11.8% ▁▁▁▄●▆▇██▆▁▃ beats 27/111 very hard · avg 4.7% 0.7% small run
ZebraLogic (maker) 37.7% ●·▁▁▁▇·▄▄▁▄▁ beats 5/24 medium · avg 43% 0.5% at its level · noisy · small run · seve…
ChineseSimpleQA (maker) 60.8% ▄··●▄▄··▄··█ beats 1/6 medium · avg 51% 0.4% at its level · small run
SimpleQA (maker) 12.2% ▄▄▄▆█●███·▂█ beats 15/32 very hard · avg 11% 0.3% noisy · small run
EuroEval HellaSwag (English), tes… 85.9% ▃·▁▁▂▁▂▅█●▄▁ beats 85/100 medium · avg 67% 0.3% sharp · noisy
EuroEval Life in the UK (English … 95.6% ▁▁▁▁▁▁▁▂▄█▄● beats 143/146 easy · avg 82% 0.1% noisy
Vectara hallucination leaderboard… 90.7% ▁▂▁▂▅●▂▃▄▃▁▁ beats 53/98 very easy · avg 90% 0.0% blunt
source results share of the score
Qwen/Qwen3-235B-A22B-Thinking-2507 mode… 24 21% ▇▇▇
Qwen/Qwen3-30B-A3B-Thinking-2507 model … 23 20% ▇▇
Qwen/Qwen3-30B-A3B-Instruct-2507 model … 25 20% ▇▇
Qwen/Qwen3-235B-A22B-Instruct-2507 mode… 11 8% ▇
Epoch AI 2 7% ▇
Aider polyglot benchmark runs 1 5% ▇
HyperCLOVA X 32B Think 3 3% ▇
Arena-Hard-Auto v2.0-Preview leaderboard 2 3% ▇
LMArena leaderboard dataset 1 2% ▇
Moonshot AI Kimi-K2-Instruct model card 1 2% ▇
LiveCodeBench official website results 1 1% ▇
BRIDGE Medical Leaderboard 1 1% ▇
EQ-Bench 3 canonical leaderboard results 1 1% ▇
WebApp1K models leaderboard 1 1% ▇
Knowledge Index of Noah's Ark 1 1% ▇
SpeakLeash CPTUB 1 1% ▇
Berkeley Function Calling Leaderboard r… 1 0% ▇
EuroEval leaderboards 2 0% ▇
THINK-Bench: Evaluating Thinking Effici… 1 0% ▇
Vectara hallucination leaderboard 1 0% ▇
prior: pull toward the middle 0% ▇
weighted average = 59.6
Licences: Qwen/Qwen3-235B-A22B-Thinking-2507 model card ([apache-2.0](../../licenses/Qwen--Qwen3-235B-A22B-Thinking-2507.txt)) · Qwen/Qwen3-30B-A3B-Thinking-2507 model card ([apache-2.0](../../licenses/Qwen--Qwen3-30B-A3B-Thinking-2507.txt)) · Qwen/Qwen3-30B-A3B-Instruct-2507 model card ([apache-2.0](../../licenses/Qwen--Qwen3-30B-A3B-Instruct-2507.txt)) · Qwen/Qwen3-235B-A22B-Instruct-2507 model card ([apache-2.0](../../licenses/Qwen--Qwen3-235B-A22B-Instruct-2507.txt)) · Epoch AI ([CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)) · Aider polyglot benchmark runs ([Apache-2.0](../../licenses/aider-polyglot-runs-Apache-2.0.txt)) · HyperCLOVA X 32B Think ([CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/)) · Arena-Hard-Auto v2.0-Preview leaderboard ([Apache-2.0](../../licenses/arena-hard-auto-Apache-2.0.txt)) · LMArena leaderboard dataset ([CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)) · Moonshot AI Kimi-K2-Instruct model card ([Modified MIT License (Kimi K2)](https://huggingface.co/moonshotai/Kimi-K2-Instruct/blob/fd1984e2b7a3350dbf7305fe73a4ede25c14de50/LICENSE)) · LiveCodeBench official website results ([CC-BY-SA-4.0](../../licenses/lcb-sharealike-NOTICE.txt)) · BRIDGE Medical Leaderboard ([MIT](../../licenses/bridge-medical-leaderboard-MIT.txt)) · EQ-Bench 3 canonical leaderboard results ([MIT](../../licenses/eqbench3-MIT.txt)) · WebApp1K models leaderboard ([MIT](../../licenses/webapp1k-leaderboard-MIT.txt)) · Knowledge Index of Noah's Ark ([CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)) · SpeakLeash CPTUB ([MIT](../../licenses/hf-leaderboard-spaces-MIT.txt)) · Berkeley Function Calling Leaderboard result archive ([Apache-2.0](../../licenses/bfcl-result-archive-v4-Apache-2.0.txt)) · EuroEval leaderboards ([MIT](https://github.com/EuroEval/leaderboards/blob/1a77f1d2f8ba1ba05ab73287fb090d76bf41b2e0/LICENSE)) · THINK-Bench: Evaluating Thinking Efficiency and Chain-of-Thought Quality of Large Reasoning Models ([CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/)) · Vectara hallucination leaderboard ([Apache-2.0](https://www.apache.org/licenses/LICENSE-2.0))
## How to run it
### Through an API
api (Alibaba) · $1.22/M blended · $0.7 in / $2.8 out per 1M tokens · [alibabacloud.com](https://www.alibabacloud.com/help/en/model-studio/models)
[price source](https://github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/alibaba/models/qwen3-235b-a22b.toml): Standard base token rate from models.dev. Provider-specific tiers, long-context surcharges, caching and promotions may differ. · [context source](https://github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/models/alibaba/qwen3-235b-a22b.toml)
### 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 235B-A22B (Apr 2025, hybrid) (ud-q2_k_xl, needs ~71 GB, from [huggingface.co/unsloth](https://huggingface.co/unsloth/Qwen3-235B-A22B-GGUF/tree/main))
```sh
llama cli -hf unsloth/Qwen3-235B-A22B-GGUF:UD-Q2_K_XL
```
3. Or start a local server and open the address it prints (http://localhost:9931)
```sh
llama serve -hf unsloth/Qwen3-235B-A22B-GGUF:UD-Q2_K_XL
```
### 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 235B-A22B (Apr 2025, hybrid) (4bit, needs ~124.9 GB, Mac only, from [huggingface.co/Qwen](https://huggingface.co/Qwen/Qwen3-235B-A22B-MLX-4bit/tree/b503f28279e60cbcbbb4281da8fa97a12c791869))
```sh
lms get https://huggingface.co/Qwen/Qwen3-235B-A22B-MLX-4bit -y
```
3. Then run lms ls to see its name, and lms chat followed by that name
Other versions: ud-q2_k_xl (~71 GB) `lms get https://huggingface.co/unsloth/Qwen3-235B-A22B-GGUF@UD-Q2_K_XL -y` · 8bit (~242.4 GB, Mac only) `lms get https://huggingface.co/Qwen/Qwen3-235B-A22B-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 235B-A22B (Apr 2025, hybrid) (Q4_K_M, needs ~141.5 GB, from [ollama.com](https://ollama.com/library/qwen3/tags))
```sh
ollama run qwen3:235b-a22b-q4_K_M
```
### 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 235B-A22B (Apr 2025, hybrid) (4bit, needs ~124.9 GB, Mac only, from [huggingface.co/Qwen](https://huggingface.co/Qwen/Qwen3-235B-A22B-MLX-4bit/tree/b503f28279e60cbcbbb4281da8fa97a12c791869))
```sh
mlx_lm.chat --model Qwen/Qwen3-235B-A22B-MLX-4bit
```
3. Or start an OpenAI-compatible server at http://localhost:8080/v1
```sh
mlx_lm.server --model Qwen/Qwen3-235B-A22B-MLX-4bit
```
Other versions: 8bit (~242.4 GB, Mac only) `mlx_lm.chat --model Qwen/Qwen3-235B-A22B-MLX-8bit`
Model repository and documentation: [huggingface.co/Qwen](https://huggingface.co/Qwen/Qwen3-235B-A22B)
## 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
FP8 W8A8 dynamic (E4M3; weigh… GSM8K 5-shot strict-match (OpenLL… 0.742 → 0.75 101.0%
FP8 W8A8 dynamic (E4M3; weigh… OpenLLM v1 average (6 tasks: MMLU… 0.738 → 0.738 100.0%
FP8 W8A8 dynamic (E4M3; weigh… MMLU 5-shot (OpenLLM v1 configura… 0.848 → 0.846 99.8%
GPTQ INT4 (w4, group_size 128… GPQA (subset not stated in card),… 0.711 → 0.719 101.1%
GPTQ INT4 (w4, group_size 128… GPQA (subset not stated in card),… 0.629 → 0.628 99.8%
GPTQ INT4 (w4, group_size 128… MMLU-Redux, non-thinking mode (Qw… 0.892 → 0.89 99.8%
GPTQ INT4 (w4, group_size 128… MMLU-Redux, thinking mode (Qwen-r… 0.927 → 0.92 99.2%
GPTQ INT4 (w4, group_size 128… LiveBench 2024-11-25, non-thinkin… 0.625 → 0.611 97.8%
GPTQ INT4 (w4, group_size 128… LiveBench 2024-11-25, thinking mo… 0.771 → 0.751 97.4%
NVFP4 W4A4 (FP4 E2M1 weights … GSM8K (OpenLLM v1 configuration, … 0.851 → 0.864 101.5%
NVFP4 W4A4 (FP4 E2M1 weights … IFEval (OpenLLM v2 leaderboard co… 0.882 → 0.882 100.0%
NVFP4 W4A4 (FP4 E2M1 weights … OpenLLM v1 average (6 tasks: arc_… 0.788 → 0.787 99.9%
NVFP4 W4A4 (FP4 E2M1 weights … OpenLLM v2 average as published (… 0.565 → 0.564 99.8%
NVFP4 W4A4 (FP4 E2M1 weights … MMLU-Pro (OpenLLM v2 leaderboard … 0.582 → 0.58 99.6%
NVFP4 W4A4 (FP4 E2M1 weights … MMLU (OpenLLM v1 configuration, 5… 0.864 → 0.858 99.3%
NVFP4 W4A4 (FP4 E2M1 weights … MATH-500 0-shot (lighteval, vLLM) 0.95 → 0.942 99.2%
NVFP4 W4A4 (FP4 E2M1 weights … Reasoning average (2 tasks: GPQA … 0.836 → 0.817 97.7%
NVFP4 W4A4 (FP4 E2M1 weights … GPQA Diamond 0-shot (lighteval, v… 0.722 → 0.692 95.8%
Each kept value links to its source.
[interactive version](https://whatmodel.sh/model/qwen3-235b-a22b) · [JSON](https://whatmodel.sh/api?format=json&all=1)