[whatmodel.sh](../../README.md) / Qwen2.5 14B Instruct
# Qwen2.5 14B Instruct
Alibaba · released 2024-09 · 14.8B params · apache-2.0
**score 52.3** ±0.8 · #332 of 732 · from 36 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 Qwen2.5 14B Instruct among every model on that task or test, worst to best.
task vs the others rank counts tests
agent tasks ▁▁▃▆●▄▅▅▄▄▂▁ #227 of 313* 6.0% 2
coding ▁▁▁▃▅▇●▇▅▄▃▁ #247 of 469 17% 7
chat & writing ▁▁▁▃▆▄●█▆▇▇▃ #255 of 402 8.8% 4
math ▁▁▃▅█●▄█▄▂▁▁ #252 of 431 13% 4
agent tasks · #227 of 313 · 6.0% of its score
Beats [o1 (2024-12-17)](o1-2024-12-17.md), [o3-mini](o3-mini.md), [DeepSeek R1 (Jan '25)](deepseek-r1.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 17.5% ▁·●█▆▅▂▃▃▂▃▁ beats 9/89 very hard · avg 12% 4.6% sharp · small run
BFCL v3 58.7% ▁▁▁▁▁▃▅█●▄▁▁ beats 78/114 medium · avg 51% 1.4% blunt · at its level
coding · #247 of 469 · 17% of its score
Beats [Qwen3-VL 32B Instruct](qwen3-vl-32b-instruct.md), [Solar Open 100B (102B-A12B)](solar-open-100b.md), [GLM-4.7-Flash](glm-4-7-flash.md) here, though they rank higher overall. Trails [QwQ-32B-Preview](qwq-32b-preview.md), [InternLM2.5 20B Chat](internlm2-5-20b-chat.md), [Ministral 3 3B Reasoning 2512](ministral-3-3b-reasoning.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
HumanEval+ 79.1% ▁·▁▁▂▃▄▄█●▄▃ beats 106/145 easy · avg 72% 5.9% sharp
HumanEval 72.1% ▁▁▁▁▂▂▆▆●▅▂▁ beats 146/205 easy · avg 71% 3.3% sharp · small run
CodeReviewQA: change localisation… 57.1% █▂▁▁▂▂▂▃▅▅●▇ beats 39/47 hard · avg 33% 2.3% sharp · at its level · noisy
HumanEval (simple-evals) 72.1% ●▅▆█▅▃··▂··▂ beats 0/19 easy · avg 76% 1.9% typical
CodeReviewQA: solution identifica… 81.9% ▃··▁▂▂▃▂█▃●▂ beats 44/47 medium · avg 51% 1.8% sharp · at its level · noisy
CodeReviewQA: automated code refi… 36.8% ▁▁▁▁▁▁▂▂▃█●▄ beats 36/47 hard · avg 32% 1.6% at its level
CodeReviewQA: change type recogni… 73.3% ▁▁▁······▁▃● beats 26/47 easy · avg 67% 0.1% blunt · at its level · noisy
chat & writing · #255 of 402 · 8.8% of its score
Beats [Mistral Small 4 (119B-A6.5B)](mistral-small-4.md), [Qwen2.5 Coder 32B Instruct](qwen2-5-coder-32b.md), [Ministral 3 14B Reasoning 2512](ministral-3-14b-reasoning.md) here, though they rank higher overall. Trails [Flan-T5 XXL](flan-t5-xxl.md), [Granite 4.0 H 1B](granite-4-0-h-1b.md), [Granite 4.0 1B](granite-4-0-1b.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
AlpacaEval 2 length-controlled 34.6% ▃▃·▁▁●▇▅·▁·▁ beats 11/22 medium · avg 39% 4.1% sharp · at its level
Arena-Hard (version/judge unspeci… 70.2% ▃▃▂▅▃▆▅●▅▅▂▃ beats 20/30 medium · avg 65% 2.9% at its level
IFEval prompt-level loose 82.4% ▁▃▁▃▁▄▁▁●▃▃▁ beats 13/22 easy · avg 78% 1.1% typical
IFEval (aggregation unspecified) 78.7% ▁▂▄▂▂●▆██▅▂▁ beats 48/145 easy · avg 76% 0.8% typical
math · #252 of 431 · 13% of its score
Beats [GPT-4o (2024-08-06)](gpt-4o-2024-08-06.md), [Command A (03-2025)](command-a.md), [GPT-4o (2024-05-13)](gpt-4o-2024-05-13.md) here, though they rank higher overall. Trails [Gemma 7B (base)](gemma-7b-base.md), [CodeGemma 7B (v1.0)](codegemma-7b.md), [Phi-2 (2.7B)](phi-2.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
MATH (subset unspecified) 70.6% ▃▃█▂▂●·····▁ beats 17/22 easy · avg 72% 4.7% small run · several runs
MGSM 79.6% ▂·▁▁▁▁▁▂▅█▃● beats 40/42 easy · avg 72% 4.3% sharp · small run
GSM8K 83.9% ▂▃··▃▂▅●▆▃▂▃ beats 12/22 very easy · avg 88% 2.5% sharp
MGSM (simple-evals) 79.6% ▁·●▁·▁▇▃█▁▃▁ beats 1/20 easy · avg 81% 1.7% typical
other tests (17)
test result vs the others in its run difficulty counts why
MMLU 81.1% ▁▂·▁·▁▁█·●▃▅ beats 15/22 easy · avg 75% 11% at its level · several runs
BIG-Bench Hard 78.4% ▁▄·▁▁·▅█·●▄▁ beats 16/22 medium · avg 72% 7.3% sharp · at its level
GPQA Diamond (simple-evals) 42.9% ▁·····●█▂▂▃▃ beats 4/22 hard · avg 39% 4.6% sharp
CPTUB average (SpeakLeash, 0-5 ju… 70.9% ▁▂▂▂▃█▃●▅▄▂▃ beats 56/95 medium · avg 65% 4.3% at its level
MMLU-Pro 63.2% ▁▁·▁▁▁▃▂●▅█▁ beats 67/188 medium · avg 58% 4.0% at its level
Polish EQ-Bench v2 (SpeakLeash) 69.6% ▁··▁▁▂▂▆▅▆●▆ beats 52/70 medium · avg 62% 3.4% at its level
AraGen-v3 3C3H score (Inception, … 36.9% ▁▁·▁▂▂▄█●▅▄▄ beats 35/66 hard · avg 29% 3.3% at its level
EvalPerf correctness (118 tasks, … 82.5% ▂·▂▂▂▆▅▆▆●▃█ beats 20/29 easy · avg 74% 3.2% typical
GPQA (subset unspecified) 42.9% ▂▁▂▂▄▃●▃▅█▃▃ beats 45/106 hard · avg 44% 2.9% sharp · at its level · small run
Open Persian LLM Leaderboard v2 a… 55.6% ▂▄▂▃█▄●▆▃▅▂▂ beats 45/81 medium · avg 57% 2.8% blunt · at its level
MMLU (simple-evals, zero-shot CoT) 79.9% ▁···●▁▅█▂▂▃▁ beats 1/22 easy · avg 80% 2.1% typical
PopQA 21.1% ▁▁▁▃▃●▁█▁▃·▄ beats 7/21 hard · avg 31% 1.9% at its level
SimpleQA 5.4% ▄▄●▆█▆███·▂█ beats 5/32 very hard · avg 4.6% 1.8% too hard for it · small run
Open PT-LLM Leaderboard average (… 79.0% ▃▁▁▁▁▁▁▁▂█●▁ beats 249/276 medium · avg 72% 1.4% at its level · noisy
Vectara hallucination leaderboard… 95.8% ▁▁▂▃▄▄●▆▃▂▂▁ beats 81/140 very easy · avg 94% 0.4% blunt · maxed out · noisy
DROP F1, 3-shot (simple-evals) 85.5% ▅▂▆█▆·●·▂▅·▂ beats 17/23 easy · avg 79% 0.3% blunt
TruthfulQA MC2 70.8% ▃▃▂·▃▃·▅█▂●▂ beats 20/21 medium · avg 60% 0.2% blunt · at its level
source results share of the score
2 OLMo 2 Furious 8 27% ▇▇▇
Microsoft Research 9 24% ▇▇▇
microsoft/phi-4 model card 6 18% ▇▇
CodeReviewQA: The Code Review Comprehen… 4 6% ▇
Reinforcing Real-world Service Agents: … 1 5% ▇
SpeakLeash CPTUB 1 4% ▇
SpeakLeash Polish EQ-Bench leaderboard 1 3% ▇
Inception Arabic/Hindi leaderboards 1 3% ▇
EvalPlus / EvalPerf published evaluatio… 1 3% ▇
Open Persian LLM Leaderboard v2 1 3% ▇
Open Portuguese LLM Leaderboard 1 1% ▇
Berkeley Function Calling Leaderboard r… 1 1% ▇
Vectara hallucination leaderboard 1 0% ▇
prior: pull toward the middle 0% ▇
weighted average = 52.3
Licences: 2 OLMo 2 Furious ([CC BY 4.0](../../licenses/paper-2501.00656v3-CC-BY-4.0.txt)) · Microsoft Research ([CC BY 4.0](../../licenses/report-2412.08905v1.txt)) · microsoft/phi-4 model card ([MIT](../../licenses/card-microsoft-phi-4.txt)) · CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models ([CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/)) · Reinforcing Real-world Service Agents: InteractCS-RL ([CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/)) · SpeakLeash CPTUB ([MIT](../../licenses/hf-leaderboard-spaces-MIT.txt)) · SpeakLeash Polish EQ-Bench leaderboard ([Apache-2.0](../../licenses/hf-leaderboard-spaces-Apache-2.0.txt)) · Inception Arabic/Hindi leaderboards ([Apache-2.0](../../licenses/hf-leaderboard-spaces-Apache-2.0.txt)) · EvalPlus / EvalPerf published evaluation runs ([Apache-2.0](../../licenses/evalplus-published-runs-Apache-2.0.txt)) · Open Persian LLM Leaderboard v2 ([Apache-2.0](../../licenses/hf-leaderboard-spaces-Apache-2.0.txt)) · Open Portuguese LLM Leaderboard ([Apache-2.0](../../licenses/hf-leaderboard-spaces-Apache-2.0.txt)) · Berkeley Function Calling Leaderboard result archive ([Apache-2.0](../../licenses/bfcl-result-archive-v4-Apache-2.0.txt)) · Vectara hallucination leaderboard ([Apache-2.0](https://www.apache.org/licenses/LICENSE-2.0))
## How to run it
### 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 Qwen2.5 14B Instruct (Q4_K_M, needs ~9 GB, from [huggingface.co/Qwen](https://huggingface.co/Qwen/Qwen2.5-14B-Instruct-GGUF/tree/b466e1f8c07172155743e8e1307507d8a4f91fbd))
```sh
llama cli -hf Qwen/Qwen2.5-14B-Instruct-GGUF:q4_k_m
```
3. Or start a local server and open the address it prints (http://localhost:9931)
```sh
llama serve -hf Qwen/Qwen2.5-14B-Instruct-GGUF:q4_k_m
```
Other versions: Q8_0 (~15.7 GB) `llama cli -hf Qwen/Qwen2.5-14B-Instruct-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 Qwen2.5 14B Instruct (Q4_K_M, needs ~9 GB, from [huggingface.co/Qwen](https://huggingface.co/Qwen/Qwen2.5-14B-Instruct-GGUF/tree/b466e1f8c07172155743e8e1307507d8a4f91fbd))
```sh
lms get https://huggingface.co/Qwen/Qwen2.5-14B-Instruct-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/Qwen2.5-14B-Instruct-GGUF@q8_0 -y` · 4bit (~8.3 GB, Mac only) `lms get https://huggingface.co/lmstudio-community/Qwen2.5-14B-Instruct-MLX-4bit -y` · 8bit (~15.7 GB, Mac only) `lms get https://huggingface.co/lmstudio-community/Qwen2.5-14B-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 Qwen2.5 14B Instruct (Q4_K_M, needs ~9.38 GB, from [ollama.com](https://ollama.com/library/qwen2.5/tags))
```sh
ollama run qwen2.5:14b
```
Other versions: Q8_0 (~18.26 GB) `ollama run qwen2.5:14b-instruct-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 Qwen2.5 14B Instruct (4bit, needs ~8.3 GB, Mac only, from [huggingface.co/lmstudio-community](https://huggingface.co/lmstudio-community/Qwen2.5-14B-Instruct-MLX-4bit/tree/e794b039d17334adc84cef973411dea4eff24a70))
```sh
mlx_lm.chat --model lmstudio-community/Qwen2.5-14B-Instruct-MLX-4bit
```
3. Or start an OpenAI-compatible server at http://localhost:8080/v1
```sh
mlx_lm.server --model lmstudio-community/Qwen2.5-14B-Instruct-MLX-4bit
```
Other versions: 8bit (~15.7 GB, Mac only) `mlx_lm.chat --model lmstudio-community/Qwen2.5-14B-Instruct-MLX-8bit`
Model repository and documentation: [huggingface.co/Qwen](https://huggingface.co/Qwen/Qwen2.5-14B-Instruct)
## 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
INT8 W8A8 (weights INT8 per-c… GSM8K — published row 'GSM-8K (5-… 0.836 → 0.855 102.4%
INT8 W8A8 (weights INT8 per-c… OpenLLM v1 average of 6 tasks (MM… 0.778 → 0.781 100.4%
INT8 W8A8 (weights INT8 per-c… MMLU — published row 'MMLU (5-sho… 0.799 → 0.797 99.8%
Each kept value links to its source.
[interactive version](https://whatmodel.sh/model/qwen2-5-14b) · [JSON](https://whatmodel.sh/api?format=json&all=1)