[whatmodel.sh](../../README.md) / Qwen3 14B
# Qwen3 14B
Alibaba · released 2025-04 · 14.8B params · apache-2.0
**score 58.1** ±1.0 · #223 of 732 · from 18 benchmark results across 13 sources
Score within the [LMArena (text)](https://whatmodel.sh/?set=lmarena-text&score=set) comparison set only.
## What it's good at
● marks Qwen3 14B among every model on that task or test, worst to best.
task vs the others rank counts tests
agent tasks ▁▁▃▆█●▅▅▄▄▂▁ #180 of 313 28% 3
math ▁▁▃▅█▆●█▄▂▁▁ #160 of 431* 9.4% 2
agent tasks · #180 of 313 · 28% 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 29.9% █··██··●█··█ beats 3/5 very hard · avg 9.9% 21% sharp · at its level
BFCL V4 (December 2025 archive) 41.0% ▁▁▁▂██●▅▆▂▂▂ beats 45/77 hard · avg 31% 5.0% at its level
BFCL v3 68.0% ▁▁▁▁▁▃▅█▅●▁▁ beats 105/114 medium · avg 51% 2.3% blunt · at its level
math · #160 of 431 · 9.4% 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
OTIS mock AIME 2024–2025 66.4% ▃▃▄▃▃▃●▇█▃▂▅ beats 94/197 very hard · avg 7.2% 7.7% sharp · at its level · noisy
GSM8K 94.7% ▃··▃··▃▃·▃·● beats 5/7 very easy · avg 93% 1.8% sharp · maxed out · noisy
other tests (13)
test result vs the others in its run difficulty counts why
GPQA Diamond 63.8% ▁▁▁▁▂▄▆●▅▇█▅ beats 91/211 hard · avg 39% 18% sharp · at its level
Medmarks-V weighted mean win rate… 53.7% ▂▃▅▂▆▃█▅●▅▃▆ beats 23/34 medium · avg 41% 8.7% blunt · at its level
BRIDGE clinical text, zero-shot a… 40.2% ▁▁▁▁▁▂▂▃▄▆●▃ beats 77/97 hard · avg 31% 6.5% blunt · at its level
AraGen-v3 3C3H score (Inception, … 29.7% ▁▁·▁▂▂▄●▄▅▄▄ beats 25/66 hard · avg 29% 6.4% at its level
CPTUB average (SpeakLeash, 0-5 ju… 70.2% ▁▂▂▂▃█▃●▅▄▂▃ beats 53/95 medium · avg 65% 6.2% typical
KINA (Knowledge Index of Noah's A… 26.0% ▁▁▂▁▃●█▆▄▅▂▃ beats 11/39 very hard · avg 21% 5.8% typical
Open Persian LLM Leaderboard v2 a… 57.8% ▂▄▂▃█▄●▆▃▅▂▂ beats 47/81 medium · avg 57% 4.4% blunt · at its level
Open PT-LLM Leaderboard average (… 79.9% ▃▁▁▁▁▁▁▁▂█●▁ beats 258/276 medium · avg 72% 2.0% at its level · noisy
EuroEval HellaSwag (English), tes… 79.8% ▃·▁▁▂▁▂▅█●▄▁ beats 72/100 medium · avg 67% 1.5% sharp · noisy
Chess Puzzles (Epoch) 4.0% ●▁▁▁▁▂▃▃▃▂▁▁ beats 42/137 very hard · avg 5.8% 1.4% sharp · too hard for it · noisy
EuroEval Life in the UK (English … 93.5% ▁▁▁▁▁▁▁▂▄█▄● beats 130/146 easy · avg 82% 0.8% noisy
Vectara hallucination leaderboard… 97.8% ▁▁▂▃▄▄█▆●▂▂▁ beats 119/140 very easy · avg 94% 0.5% blunt · maxed out · noisy
Vectara hallucination leaderboard… 94.6% ▁▂▁▂▅█▂▃●▃▁▁ beats 84/98 very easy · avg 90% 0.0% blunt
source results share of the score
Epoch AI 3 27% ▇▇▇
MUA-RL: Multi-turn User-interacting Age… 1 21% ▇▇▇
Medmarks: A Comprehensive Open-Source L… 1 9% ▇
Berkeley Function Calling Leaderboard r… 2 7% ▇
BRIDGE Medical Leaderboard 1 7% ▇
Inception Arabic/Hindi leaderboards 1 6% ▇
SpeakLeash CPTUB 1 6% ▇
Knowledge Index of Noah's Ark 1 6% ▇
Open Persian LLM Leaderboard v2 1 4% ▇
EuroEval leaderboards 2 2% ▇
Open Portuguese LLM Leaderboard 1 2% ▇
Unsupervised Confidence Calibration for… 1 2% ▇
Vectara hallucination leaderboard 2 0% ▇
prior: pull toward the middle 0% ▇
weighted average = 58.1
Licences: Epoch AI ([CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)) · MUA-RL: Multi-turn User-interacting Agent Reinforcement Learning for agentic tool use ([CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/)) · Medmarks: A Comprehensive Open-Source LLM Benchmark Suite for Medical Tasks ([CC BY 4.0](../../licenses/paper-2605.01417v1-CC-BY-4.0.txt)) · Berkeley Function Calling Leaderboard result archive ([Apache-2.0](../../licenses/bfcl-result-archive-v4-Apache-2.0.txt)) · BRIDGE Medical Leaderboard ([MIT](../../licenses/bridge-medical-leaderboard-MIT.txt)) · Inception Arabic/Hindi leaderboards ([Apache-2.0](../../licenses/hf-leaderboard-spaces-Apache-2.0.txt)) · SpeakLeash CPTUB ([MIT](../../licenses/hf-leaderboard-spaces-MIT.txt)) · Knowledge Index of Noah's Ark ([CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)) · Open Persian LLM Leaderboard v2 ([Apache-2.0](../../licenses/hf-leaderboard-spaces-Apache-2.0.txt)) · EuroEval leaderboards ([MIT](https://github.com/EuroEval/leaderboards/blob/1a77f1d2f8ba1ba05ab73287fb090d76bf41b2e0/LICENSE)) · Open Portuguese LLM Leaderboard ([Apache-2.0](../../licenses/hf-leaderboard-spaces-Apache-2.0.txt)) · Unsupervised Confidence Calibration for Reasoning LLMs from a Single Generation ([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) · $0.61/M blended · $0.35 in / $1.4 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/qwen3-14b.toml): Standard base token rate from models.dev. Provider-specific tiers, long-context surcharges, caching and promotions may differ. · [context source](https://huggingface.co/Qwen/Qwen3-14B/blob/40c069824f4251a91eefaf281ebe4c544efd3e18/README.md): Native/default model-card context. Any YaRN extension must be configured separately; hosted provider limits may differ.
### 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 14B (Q4_K_M, needs ~9 GB, from [huggingface.co/Qwen](https://huggingface.co/Qwen/Qwen3-14B-GGUF/tree/530227a7d994db8eca5ab5ced2fb692b614357fd))
```sh
llama cli -hf Qwen/Qwen3-14B-GGUF:Q4_K_M
```
3. Or start a local server and open the address it prints (http://localhost:9931)
```sh
llama serve -hf Qwen/Qwen3-14B-GGUF:Q4_K_M
```
Other versions: Q8_0 (~15.7 GB) `llama cli -hf Qwen/Qwen3-14B-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 14B (Q4_K_M, needs ~9 GB, from [huggingface.co/Qwen](https://huggingface.co/Qwen/Qwen3-14B-GGUF/tree/530227a7d994db8eca5ab5ced2fb692b614357fd))
```sh
lms get https://huggingface.co/Qwen/Qwen3-14B-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 (~15.7 GB) `lms get https://huggingface.co/Qwen/Qwen3-14B-GGUF@Q8_0 -y` · 4bit (~8.3 GB, Mac only) `lms get https://huggingface.co/lmstudio-community/Qwen3-14B-MLX-4bit -y` · 8bit (~15.7 GB, Mac only) `lms get https://huggingface.co/lmstudio-community/Qwen3-14B-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 14B (Q4_K_M, needs ~9.38 GB, from [ollama.com](https://ollama.com/library/qwen3/tags))
```sh
ollama run qwen3:14b
```
Other versions: Q8_0 (~18.26 GB) `ollama run qwen3:14b-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 14B (4bit, needs ~8.3 GB, Mac only, from [huggingface.co/lmstudio-community](https://huggingface.co/lmstudio-community/Qwen3-14B-MLX-4bit/tree/b5d17e319ff9734f059b42b8b1f0834932bbb12c))
```sh
mlx_lm.chat --model lmstudio-community/Qwen3-14B-MLX-4bit
```
3. Or start an OpenAI-compatible server at http://localhost:8080/v1
```sh
mlx_lm.server --model lmstudio-community/Qwen3-14B-MLX-4bit
```
Other versions: 8bit (~15.7 GB, Mac only) `mlx_lm.chat --model lmstudio-community/Qwen3-14B-MLX-8bit`
Model repository and documentation: [huggingface.co/Qwen](https://huggingface.co/Qwen/Qwen3-14B)
## 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
AWQ INT4 (w4, group_size 128,… MMLU-Redux, thinking mode (Qwen-r… 0.886 → 0.885 99.9%
AWQ INT4 (w4, group_size 128,… MMLU-Redux, non-thinking mode (Qw… 0.82 → 0.815 99.4%
AWQ INT4 (w4, group_size 128,… LiveBench 2024-11-25, thinking mo… 0.713 → 0.7 98.2%
AWQ INT4 (w4, group_size 128,… GPQA (subset not stated in card),… 0.548 → 0.538 98.2%
AWQ INT4 (w4, group_size 128,… AIME 2024, thinking mode (Qwen-re… 0.793 → 0.77 97.1%
AWQ INT4 (w4, group_size 128,… GPQA (subset not stated in card),… 0.64 → 0.621 97.0%
AWQ INT4 (w4, group_size 128,… LiveBench 2024-11-25, non-thinkin… 0.596 → 0.574 96.3%
FP8 block (weights FP8 128x12… OpenLLM v1 average of 6 tasks (AR… 0.748 → 0.749 100.2%
FP8 block (weights FP8 128x12… GSM8K — published row 'GSM8K (Str… 0.884 → 0.884 100.0%
FP8 block (weights FP8 128x12… MMLU — published row 'MMLU (Acc, … 0.788 → 0.787 99.8%
FP8 block (weights FP8 128x12… OpenLLM v2 average of 6 tasks (IF… 0.315 → 0.311 98.8%
FP8 dynamic (weights FP8 per-… GPQA Diamond — published row 'GPQ… 0.616 → 0.641 104.1%
FP8 dynamic (weights FP8 per-… AIME 2025 — published row 'AIME 2… 0.663 → 0.67 100.9%
FP8 dynamic (weights FP8 per-… AIME 2024 — published row 'AIME 2… 0.766 → 0.769 100.4%
FP8 dynamic (weights FP8 per-… OpenLLM v1 average of 6 tasks (MM… 0.632 → 0.634 100.3%
FP8 dynamic (weights FP8 per-… OpenLLM v2 average of 6 tasks (MM… 0.388 → 0.388 100.1%
FP8 dynamic (weights FP8 per-… MMLU — published row 'MMLU (5-sho… 0.768 → 0.765 99.6%
FP8 dynamic (weights FP8 per-… LiveCodeBench — published row 'Li… 0.608 → 0.606 99.5%
FP8 dynamic (weights FP8 per-… GSM8K — published row 'GSM-8K (5-… 0.676 → 0.662 97.9%
INT4 W4A16 (weight-only INT4,… GSM8K — published row 'GSM-8K (5-… 0.676 → 0.716 105.8%
INT4 W4A16 (weight-only INT4,… OpenLLM v1 average of 6 tasks (MM… 0.632 → 0.644 102.0%
INT4 W4A16 (weight-only INT4,… AIME 2024 — published row 'AIME 2… 0.766 → 0.779 101.8%
INT4 W4A16 (weight-only INT4,… OpenLLM v2 average of 6 tasks (MM… 0.388 → 0.386 99.5%
INT4 W4A16 (weight-only INT4,… GPQA Diamond — published row 'GPQ… 0.616 → 0.611 99.2%
INT4 W4A16 (weight-only INT4,… MMLU — published row 'MMLU (5-sho… 0.768 → 0.759 98.8%
INT4 W4A16 (weight-only INT4,… AIME 2025 — published row 'AIME 2… 0.663 → 0.65 98.0%
INT4 W4A16 (weight-only INT4,… LiveCodeBench — published row 'Li… 0.608 → 0.586 96.3%
NVFP4 W4A4 (FP4 weights and a… GSM8K — published row 'gsm8k' (Op… 0.887 → 0.882 99.5%
NVFP4 W4A4 (FP4 weights and a… OpenLLM v1 average of 6 tasks (ar… 0.745 → 0.739 99.2%
NVFP4 W4A4 (FP4 weights and a… MMLU — published row 'mmlu' (Open… 0.789 → 0.772 97.9%
NVFP4 W4A4 (FP4 weights and a… OpenLLM v2 average of 6 tasks (BB… 0.528 → 0.51 96.7%
NVFP4 W4A4 (FP4 weights and a… AIME 2025 — published row 'AIME25… 0.69 → 0.655 95.0%
NVFP4 W4A4 (FP4 weights and a… GPQA Diamond — published row 'GPQ… 0.65 → 0.604 93.0%
NVFP4 W4A4 (FP4 weights and a… Reasoning average of 3 tasks (AIM… 0.699 → 0.638 91.2%
NVFP4 W4A4 (FP4 weights and a… AIME 2024 — published row 'AIME24… 0.759 → 0.655 86.4%
Each kept value links to its source.
[interactive version](https://whatmodel.sh/model/qwen3-14b) · [JSON](https://whatmodel.sh/api?format=json&all=1)