[whatmodel.sh](../../README.md) / Qwen3-Next 80B-A3B Instruct
# Qwen3-Next 80B-A3B Instruct
Alibaba · released 2025-09 · 80B params (3B active) · apache-2.0
**score 62.9** ±0.8 · #172 of 732 · from 31 benchmark results across 6 sources
Score within the [LMArena (text)](https://whatmodel.sh/?set=lmarena-text&score=set) comparison set only.
## What it's good at
● marks Qwen3-Next 80B-A3B Instruct among every model on that task or test, worst to best.
task vs the others rank counts tests
agent tasks ▁▁▃▆●▄▅▅▄▄▂▁ #205 of 313 15% 6
coding ▁▁▁▃▅▇█●▅▄▃▁ #177 of 469 16% 3
chat & writing ▁▁▁▃▆▄▆█▆●▇▃ #117 of 402 10% 3
math ▁▁▃▅█▆▄●▄▂▁▁ #78 of 431 8.0% 3
agent tasks · #205 of 313 · 15% 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 (maker) 13.2% ▁·●█▆▅▂▃▃▂▃▁ beats 3/89 very hard · avg 5.3% 4.0% sharp · at its level
τ-bench Retail (maker) 60.9% ▁····▃·▂●█▂▆ beats 11/28 hard · avg 32% 3.3% at its level
τ²-bench Retail (maker) 57.3% ▁····▂▁●▆█▃▁ beats 9/44 medium · avg 35% 2.7% at its level
τ²-bench Airline (maker) 45.5% ▁▁·▂▁●▂▁█▅▂▃ beats 9/42 hard · avg 22% 2.2% at its level
τ-bench Airline (maker) 44.0% ▁▁▁·▅▄▁●▇▁▁▇ beats 12/28 hard · avg 19% 2.1% at its level
BFCL v3 (maker) 70.3% ▁▁▁▁▁▃▅█▆▅●▁ beats 151/171 medium · avg 56% 1.0% blunt · at its level
coding · #177 of 469 · 16% of its score
Beats [Qwen3-Coder-Next (80B-A3B)](qwen3-coder-next.md), [Qwen3-VL 235B-A22B Instruct](qwen3-vl-235b-a22b-instruct.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… 49.8% ▂▁▁▃▆▃▄●█▃▄▂ beats 41/71 very hard · avg 8.2% 11% sharp · at its level
LiveCodeBench v6 (Feb–May 2025) (… 56.6% ▁▁▃▁█●▄▇▅▅▃█ beats 11/38 hard · avg 27% 2.4% at its level
MultiPL-E (maker) 87.8% ▃··▄·▅▅▇▄█▃● beats 30/31 easy · avg 68% 2.0% typical
chat & writing · #117 of 402 · 10% of its score
Beats [LongCat 2.0 (1.6T-A48B)](longcat-2-0.md), [Step 3.5 Flash (196B-A11B)](step-3-5-flash.md), [GPT-5 mini](gpt-5-mini.md) here, though they rank higher overall. Trails [Nemotron-Cascade-8B](nemotron-cascade-8b.md), [Mistral Large 3 (675B-A41B)](mistral-large-3.md), [Gemini 2.5 Flash](gemini-2-5-flash.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
LMArena (text) 1398.5 ▁▁▁▂▃▃▃▆▆●█▃ beats 244/385 avg 1260 6.1% blunt
Arena-Hard v2 aggregate (maker) 82.7% ▁▁··▄▃▁▃█▁▁● beats 21/22 hard · avg 32% 2.3% sharp · at its level
IFEval (aggregation unspecified) … 87.6% ▁▂▄▂▂▄▆●█▅▂▁ beats 89/145 easy · avg 76% 1.6% typical
math · #78 of 431 · 8.0% of its score
Beats [Claude Sonnet 4.6](claude-sonnet-4-6.md), [Claude Opus 4.5](claude-opus-4-5.md), [Qwen3.7 Plus](qwen3-7-plus.md) here, though they rank higher overall. Trails [Granite 4.2 30B](granite-4-2-30b.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
HMMT 2025 (edition unspecified) (… 54.1% ▃█●▅▇▅▁·▁··▁ beats 10/32 very hard · avg 15% 3.4% sharp · at its level
AIME 2025 (maker) 69.5% ▁·▁▁▁▂▅▆●▇▅▁ beats 63/133 hard · avg 25% 2.8% sharp · at its level
PolyMATH (maker) 45.9% ▁···▃▁·●█▄▄▄ beats 6/23 hard · avg 24% 1.8% at its level
other tests (16)
test result vs the others in its run difficulty counts why
Medmarks-V mean win rate (19 veri… 0.5687 ▁▄▃▃▇▄█▃●▁▁▁ beats 25/30 avg 0.418 11% sharp
GPQA (subset unspecified) (maker) 72.9% ▂▁▂▂▄▃▆▃▅●▃▃ beats 82/106 hard · avg 44% 5.6% sharp · at its level
Medmarks-V weighted mean win rate… 56.9% ▂▃▅▂▆▃█▅█●▃▆ beats 26/34 medium · avg 41% 4.9% blunt · at its level
KINA (Knowledge Index of Noah's A… 30.1% ▁▁▂▁▃▄█●▄▅▂▃ beats 21/39 very hard · avg 21% 3.9% typical
BRIDGE clinical text, zero-shot a… 39.8% ▁▁▁▁▁▂▂▃▄▆●▃ beats 71/97 hard · avg 31% 3.9% blunt · at its level
MMLU-Pro (maker) 80.6% ▁▁·▁▁▁▃▂▄▅●▁ beats 129/188 medium · avg 58% 3.5% typical
INCLUDE (maker) 78.9% ▂····▂▂▄▄▄●▄ beats 25/36 medium · avg 57% 3.1% at its level
MMLU-Redux (maker) 90.9% ▁▁▃▁▂▃▄▅▇●█▄ beats 41/63 easy · avg 78% 2.9% typical
Creative Writing v3 (maker) 85.3% ▂····▂▆··▆●▆ beats 8/14 easy · avg 69% 2.1% typical
SuperGPQA (maker) 58.8% ▂▁▂▂▅▆▃▇▇●█▃ beats 32/46 medium · avg 36% 2.0% at its level
MMLU-ProX (maker) 76.7% ▁·▁▁▁▂▁▃●█▅▄ beats 14/36 medium · avg 57% 2.0% typical
LiveBench 2024-11-25 (maker) 75.8% ▄··▂·▂·▂●▆▂▂ beats 8/13 medium · avg 55% 1.5% at its level
WritingBench (maker) 87.3% ▁·▁▂▁··▅█▂●▁ beats 19/20 easy · avg 70% 1.5% typical
Multi-IF (aggregation unspecified… 75.8% ▁▁▁▁▁▁▃▄●▄▁▂ beats 19/41 medium · avg 59% 1.4% typical
EuroEval HellaSwag (English), tes… 90.5% ▃·▁▁▂▁▂▅█▇●▁ beats 93/100 medium · avg 67% 0.6% sharp · noisy
EuroEval Life in the UK (English … 94.4% ▁▁▁▁▁▁▁▂▄█▄● beats 138/146 easy · avg 82% 0.3% noisy
source results share of the score
Qwen/Qwen3-Next-80B-A3B-Instruct model … 24 69% ▇▇▇▇▇▇▇▇
Medmarks: A Comprehensive Open-Source L… 2 16% ▇▇
LMArena leaderboard dataset 1 6% ▇
Knowledge Index of Noah's Ark 1 4% ▇
BRIDGE Medical Leaderboard 1 4% ▇
EuroEval leaderboards 2 1% ▇
prior: pull toward the middle 0% ▇
weighted average = 62.9
Licences: Qwen/Qwen3-Next-80B-A3B-Instruct model card ([apache-2.0](../../licenses/Qwen--Qwen3-Next-80B-A3B-Instruct.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)) · LMArena leaderboard dataset ([CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)) · Knowledge Index of Noah's Ark ([CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)) · BRIDGE Medical Leaderboard ([MIT](../../licenses/bridge-medical-leaderboard-MIT.txt)) · EuroEval leaderboards ([MIT](https://github.com/EuroEval/leaderboards/blob/1a77f1d2f8ba1ba05ab73287fb090d76bf41b2e0/LICENSE))
## How to run it
### Through an API
api (Deep Infra) · $0.34/M blended · $0.09 in / $1.1 out per 1M tokens · [deepinfra.com](https://deepinfra.com/models)
[price source](https://github.com/anomalyco/models.dev/blob/a668908cfe7073faa0363699486fc41a3ff52412/providers/deepinfra/models/Qwen/Qwen3-Next-80B-A3B-Instruct.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-next-80b-a3b-instruct.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-Next 80B-A3B Instruct (ud-q2_k_xl, needs ~24.5 GB, from [huggingface.co/unsloth](https://huggingface.co/unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF/tree/main))
```sh
llama cli -hf unsloth/Qwen3-Next-80B-A3B-Instruct-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-Next-80B-A3B-Instruct-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-Next 80B-A3B Instruct (4bit, needs ~44.8 GB, Mac only, from [huggingface.co/lmstudio-community](https://huggingface.co/lmstudio-community/Qwen3-Next-80B-A3B-Instruct-MLX-4bit/tree/6d14222d4664dba1aa9f3251696c8254c6e994af))
```sh
lms get https://huggingface.co/lmstudio-community/Qwen3-Next-80B-A3B-Instruct-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 (~24.5 GB) `lms get https://huggingface.co/unsloth/Qwen3-Next-80B-A3B-Instruct-GGUF@UD-Q2_K_XL -y` · 8bit (~84.7 GB, Mac only) `lms get https://huggingface.co/lmstudio-community/Qwen3-Next-80B-A3B-Instruct-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-Next 80B-A3B Instruct (Q4_K_M, needs ~48.5 GB, from [ollama.com](https://ollama.com/library/qwen3-next/tags))
```sh
ollama run qwen3-next:80b-a3b-instruct-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-Next 80B-A3B Instruct (4bit, needs ~44.8 GB, Mac only, from [huggingface.co/lmstudio-community](https://huggingface.co/lmstudio-community/Qwen3-Next-80B-A3B-Instruct-MLX-4bit/tree/6d14222d4664dba1aa9f3251696c8254c6e994af))
```sh
mlx_lm.chat --model lmstudio-community/Qwen3-Next-80B-A3B-Instruct-MLX-4bit
```
3. Or start an OpenAI-compatible server at http://localhost:8080/v1
```sh
mlx_lm.server --model lmstudio-community/Qwen3-Next-80B-A3B-Instruct-MLX-4bit
```
Other versions: 8bit (~84.7 GB, Mac only) `mlx_lm.chat --model lmstudio-community/Qwen3-Next-80B-A3B-Instruct-MLX-8bit`
Model repository and documentation: [huggingface.co/Qwen](https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Instruct)
## Published speed measurements
Reference results for the hardware, engine and settings below. These are not verified speeds of the downloads above or predictions for your machine.
hardware precision speed engine and settings
AMD Ryzen AI Max+ 395 (Stri… UD-Q4_K_XL 62.09 tok/s llama.cpp c841aee (build_number 1…
AMD Ryzen AI Max+ 395 (Stri… UD-Q4_K_XL 59.06 tok/s llama.cpp 1348f67c5 (build_number…
AMD Ryzen AI Max+ 395 (Stri… UD-Q4_K_XL 49.57 tok/s llama.cpp e77056f (build_number 1…
AMD Ryzen AI Max+ 395 (Stri… UD-Q4_K_XL 54.92 tok/s llama.cpp b8933, Vulkan (RADV) · …
NVIDIA GeForce RTX 5090 GGUF Q2_K_XL (U… 76.1 tok/s llama.cpp (GGUF; version not stat…
NVIDIA GeForce RTX 5090 GGUF Q4_K_M (Un… 4.7 tok/s llama.cpp (GGUF; version not stat…
Apple M4 Pro (14-core CPU /… MLX 4-bit 52.3 tok/s MLX v0.30.6 (macOS 26 Tahoe) · Of…
Apple M3 Ultra (28-core CPU… MLX 4-bit 49.1 tok/s MLX v0.30.6 (macOS 26 Tahoe) · Of…
Each speed links to its source, which has the full settings.
## 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 block (E4M3; weights stat… GSM8K 5-shot strict-match (OpenLL… 0.31 → 0.32 103.2%
FP8 block (E4M3; weights stat… OpenLLM v2 average (6 tasks: IFEv… 0.57 → 0.575 100.8%
FP8 block (E4M3; weights stat… OpenLLM v1 average (6 tasks: ARC-… 0.691 → 0.694 100.4%
FP8 block (E4M3; weights stat… IFEval inst-level strict acc 0-sh… 0.904 → 0.905 100.1%
FP8 block (E4M3; weights stat… MMLU 5-shot acc (OpenLLM v1 confi… 0.856 → 0.856 100.0%
FP8 block (E4M3; weights stat… MMLU-Pro 5-shot acc (OpenLLM v2 l… 0.597 → 0.597 99.9%
FP8 W8A8 dynamic (E4M3; weigh… GSM8K 5-shot strict-match (OpenLL… 0.31 → 0.332 107.1%
FP8 W8A8 dynamic (E4M3; weigh… OpenLLM v1 average (6 tasks: ARC-… 0.691 → 0.696 100.7%
FP8 W8A8 dynamic (E4M3; weigh… MMLU-Pro 5-shot acc (OpenLLM v2 l… 0.597 → 0.6 100.5%
FP8 W8A8 dynamic (E4M3; weigh… OpenLLM v2 average (6 tasks: IFEv… 0.57 → 0.573 100.4%
FP8 W8A8 dynamic (E4M3; weigh… IFEval inst-level strict acc 0-sh… 0.904 → 0.903 99.9%
FP8 W8A8 dynamic (E4M3; weigh… MMLU 5-shot acc (OpenLLM v1 confi… 0.856 → 0.854 99.8%
INT4 W4A16 (GPTQ, symmetric, … IFEval inst-level strict acc 0-sh… 0.775 → 0.807 104.2%
INT4 W4A16 (GPTQ, symmetric, … GSM8K 5-shot strict-match (OpenLL… 0.816 → 0.822 100.7%
INT4 W4A16 (GPTQ, symmetric, … OpenLLM v2 average (6 tasks: IFEv… 0.577 → 0.579 100.5%
INT4 W4A16 (GPTQ, symmetric, … OpenLLM v1 average (6 tasks: ARC-… 0.739 → 0.738 99.8%
INT4 W4A16 (GPTQ, symmetric, … MMLU 5-shot acc (OpenLLM v1 confi… 0.856 → 0.85 99.4%
INT4 W4A16 (GPTQ, symmetric, … MMLU-Pro 5-shot acc (OpenLLM v2 l… 0.763 → 0.755 98.9%
INT8 W8A8 (weights static per… AIME 2025 0-shot (lighteval aime2… 0.628 → 0.65 103.5%
INT8 W8A8 (weights static per… MATH-500 0-shot (lighteval math_5… 0.897 → 0.906 100.9%
INT8 W8A8 (weights static per… GPQA Diamond 0-shot (lighteval gp… 0.746 → 0.752 100.8%
INT8 W8A8 (weights static per… MMLU-Pro 5-shot (lm-evaluation-ha… 0.786 → 0.788 100.3%
INT8 W8A8 (weights static per… IFEval (lm-evaluation-harness ife… 0.914 → 0.915 100.1%
INT8 W8A8 (weights static per… GSM8K 5-shot (lm-evaluation-harne… 0.697 → 0.697 100.0%
Each kept value links to its source.
[interactive version](https://whatmodel.sh/model/qwen3-next-80b-a3b-instruct) · [JSON](https://whatmodel.sh/api?format=json&all=1)