whatmodel.sh / Qwen2.5 3B Instruct

Qwen2.5 3B Instruct

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Qwen2.5 3B Instruct is an open-weights language model from Alibaba, released September 2024. Its published size is 3.09B parameters and a 33k-token context window. Licence: qwen-research.

Catalogue identity notes

Maker repo Qwen/Qwen2.5-3B-Instruct @aa8e725: licence tag other (qwen-research). 3,085,938,688 parameters from the Hugging Face safetensors metadata. Dense, so active = total. Released 2024-09: month the maker's Hugging Face repo was created (createdAt 2024-09-17). Ollama tags qwen2.5:3b-instruct-q4_K_M, qwen2.5:3b-instruct-q8_0 are the library's instruct/chat builds of this checkpoint (tag names on the pinned tags page); download sizes are estimated from the parameter count.

42.7±0.6#490 of 719 LLMsfrom 32 benchmark results across 11 sources

$ how it compares

The LLMs ranked either side of Qwen2.5 3B Instruct, out of 828 with a score, then more from Alibaba.

#modelscorewhereRAMreleased
578Granite 3.1 8B Instruct IBM43±1local~6.6GB2024-12
579Llama 3.2 11B Vision Instruct Meta43±2local~8.8GB2024-09
580Qwen2.5 3B Instruct Alibaba43±1local~3.3GB2024-09
581DeepSeek-Coder 1.3B Instruct DeepSeek43±1local~2.4GB2023-10
582Aya 23 8B Cohere For AI43±3local~6.2GB2024-05
more from Alibaba
18Qwen3.8 Max (2.4T-A95B) Alibaba78±1api $3.00/M +local~380GB2026-08
34Qwen3.8 Max (0902) Alibaba76±1api $3.00/M2026-09
37Qwen3.8-Flash-Next Alibaba75±1local~59GB2026-08

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #254 of 300 * · 0.8% of its score · 1 test

Beats DeepSeek R1 0528 Qwen3 8B, Magistral Small 1.0 (2506), Llama 4 Scout (109B-A17B) here, though they rank higher overall. Trails Granite 4.0 H 1B, Granite 4.0 1B here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
BFCL v350.4% ✎▁▁▁▁▁▃▅█▅▄▁▁ beats 35/114medium · avg 52%▇ 0.8%blunt · at its level
coding▁▁▁▃▆██▇▆▄▃▁ #322 of 458 · 12% of its score · 5 tests

Beats GLM-4.7-Flash, Qwen3-VL 8B Instruct, Qwen3-VL 4B Instruct here, though they rank higher overall. Trails Phi-3-mini-128k-instruct (Apr 2024), DeepSeek LLM 67B Chat, WizardLM 70B V1.0 here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
CRUXEval (ZeroEval, 800 problems)33.1% ✎▁▃▆▃▆█▆▃▂·▁▁ beats 1/47medium · avg 61%▇▇▇▇ 10%sharp · at its level
CodeReviewQA: solution identification (hard), invariant accuracy45.2% ✎▃··▁▂▂▃▂█▃▄▂ beats 21/47medium · avg 54%▇ 0.9%sharp · at its level · noisy
CodeReviewQA: change localisation (hard), invariant accuracy35.0% ✎█▂▁▁▂▂▂▃▅▅▅▇ beats 29/47medium · avg 36%▇ 0.7%sharp · noisy
CodeReviewQA: automated code refinement, exact match rate21.3% ✎▁▁▁▁▁▁▂▃▃█▅▄ beats 13/47hard · avg 33%▇ 0.6%typical
CodeReviewQA: change type recognition, invariant accuracy68.8% ✎▁▁▁······▁▃█ beats 17/47easy · avg 67%▇ 0.0%blunt · at its level · noisy
chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #273 of 393 * · 2.5% of its score · 1 test

Beats Ministral 3 14B Reasoning 2512, Ministral 3 14B, Gemma 4 E4B here, though they rank higher overall. Trails Flan-T5 XXL, EXAONE 3.5 2.4B Instruct, LFM2 1.2B here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
Arabic IFEval prompt-level (Inception)38.1% ✎▁·▁▁▂▁▄▄▆▄█▄ beats 21/38medium · avg 49%▇ 2.5%sharp · at its level
math▁▂▃▅▆▅▄█▄▁▁▁ #317 of 398 · 16% of its score · 4 tests

Beats Athene-70B, OLMo 2 32B Instruct (0325), Gemma 2 9B IT here, though they rank higher overall. Trails Gemma 7B (base), DeepSeek R1 Distill Qwen 1.5B, Tulu 2 DPO 13B here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
GSM8K80.6% ✎█·█▄▄█···█·▄ beats 3/10very easy · avg 91%▇▇ 6.1%sharp
MGSM53.5% ✎▅▃▃█·▅···▃·▃ beats 3/10easy · avg 71%▇▇ 4.3%at its level
MATH Level 5 (ZeroEval, 721 problems)25.5% ✎▁▁▁▆█▂▇▇▄··▂ beats 14/38hard · avg 34%▇ 2.9%sharp
MATH61.7% ✎█▄██···▄▄▄·▄ beats 8/10medium · avg 64%▇ 2.6%at its level
other tests21 not in a task · 69% of its score
testresultvs the others in its rundifficultycountswhy
MMLU-Redux (ZeroEval, 2,778 questions)64.3% ✎▁·▁▁▄▄▄▅█▄▂▁ beats 6/53easy · avg 77%▇▇▇▇▇▇▇ 18%at its level
ARC Challenge82.6% ✎▂▂▂█▂···▄··▂ beats 3/10very easy · avg 91%▇▇▇ 8.6%sharp
MMLU65.0% ✎█▄█▄▄▄·▄▄··▄ beats 4/10easy · avg 77%▇▇ 5.3%at its level
OpenBookQA79.3% ✎▂··▄█▂·▂···▄ beats 2/10easy · avg 87%▇▇ 4.9%at its level
BIG-Bench Hard56.2% ✎██▄·█▄·▄▄··▄ beats 4/10easy · avg 74%▇▇ 4.9%sharp · at its level
MMLU-Pro44.7% ✎█▄▄·█▄▄▄·▄·▄ beats 5/10medium · avg 57%▇ 3.2%at its level
Multilingual MMLU (Microsoft)55.9% ✎█▄▄██··█···▄ beats 6/10easy · avg 65%▇ 3.0%at its level
CPTUB average (SpeakLeash, 0-5 judge scale)50.1% ✎▁▂▂▃▃█▃▆▅▄▂▃ beats 16/95easy · avg 66%▇ 2.6%at its level
13 more tests
testresultvs the others in its rundifficultycountswhy
ZebraLogic (ZeroEval, 1,000 puzzles)4.8% ✎▁▃▃▆██▁▁·▁▁▁ beats 2/57hard · avg 18%▇ 2.4%sharp · too hard for it
Polish EQ-Bench v2 (SpeakLeash)36.1% ✎▁··▁▁▂▃▆▅▆█▆ beats 9/70medium · avg 64%▇ 2.1%at its level
WinoGrande63.3% ✎▃····▃▃█▃·▅▅ beats 3/10medium · avg 74%▇ 1.9%at its level
HellaSwag74.6% ✎▃▅█·▃▃▃···▃▃ beats 5/10easy · avg 84%▇ 1.9%at its level
BoolQ65.4% ✎▃··▃··██▃·▃▃ beats 0/10easy · avg 85%▇ 1.7%at its level
Social IQA75.3% ✎▃·▃▅▅█·▃···▃ beats 8/10easy · avg 78%▇ 1.6%blunt · at its level
BRIDGE clinical text, zero-shot average (87 tasks)26.6% ✎▁▁▁▁▁▂▃▃▄▆█▃ beats 23/97hard · avg 32%▇ 1.5%blunt · at its level
Open Persian LLM Leaderboard v2 average (21 tasks)43.5% ✎▂▄▂▃█▄▃▆▃▅▂▂ beats 16/81medium · avg 58%▇ 1.5%blunt · at its level
Open PT-LLM Leaderboard average (9 Portuguese tasks)70.8% ✎▃▁▁▁▁▁▁▁▂█▅▁ beats 174/276medium · avg 73%▇ 0.8%at its level · noisy
GPQA (subset unspecified)23.4% ✎█··▄▂··▂▂··▄ beats 1/10very hard · avg 33%▇ 0.8%sharp · too hard for it · noisy
TruthfulQA MC264.3% ✎▂·█▂▂▄····▂▂ beats 4/10easy · avg 69%▇ 0.5%blunt · at its level
PIQA72.6% ✎▃·▅·▃█·▃▅··▃ beats 1/10easy · avg 83%▇ 0.5%blunt · at its level
Vectara hallucination leaderboard, 2025 dataset (HHEM-2.3, factual consistency rate)93.0% ✎▁▁▂▃▄▄█▆▃▂▂▁ beats 42/140very easy · avg 94%▇ 0.3%blunt · noisy

$ where to run it

llama.cpp · recommended

version

  1. Install llama.cpp · Homebrew and winget installs name the commands llama-cli and llama-server instead.
    macOS / Linux
    curl -LsSf https://llama.app/install.sh | sh
    Windows
    irm https://llama.app/install.ps1 | iex

    other ways to install ↗

  2. Run Qwen2.5 3B Instruct · needs ~2.1 GB · from huggingface.co/Qwen (official)
    chat in the terminal
    llama cli -hf Qwen/Qwen2.5-3B-Instruct-GGUF:q4_k_m
  3. Or start a local server and open the address it prints (http://localhost:9931)
    llama serve -hf Qwen/Qwen2.5-3B-Instruct-GGUF:q4_k_m

LM Studio

version

  1. Install LM Studio · Open the app once before using the lms command.

    download it (macOS / Windows / Linux) ↗

  2. Run Qwen2.5 3B Instruct · needs ~2.1 GB · from huggingface.co/Qwen (official)

    open in LM Studio ↗

  3. Or download it in the terminal
    lms get https://huggingface.co/Qwen/Qwen2.5-3B-Instruct-GGUF@q4_k_m -y
  4. Then run lms ls to see its name, and lms chat followed by that name

Ollama

version

  1. Install Ollama

    download it (macOS / Windows) ↗

    Linux
    curl -fsSL https://ollama.com/install.sh | sh
  2. Run Qwen2.5 3B Instruct · needs ~2.35 GB · from ollama.com
    chat in the terminal
    ollama run qwen2.5:3b-instruct-q4_K_M

MLX (Mac)

version

  1. Install MLX (Mac) · Apple silicon Macs only, text only.
    with uv
    uv tool install mlx-lm
    or
    pip install mlx-lm
  2. Run Qwen2.5 3B Instruct · needs ~1.7 GB, Mac only · from huggingface.co/lmstudio-community
    chat in the terminal
    mlx_lm.chat --model lmstudio-community/Qwen2.5-3B-Instruct-MLX-4bit
  3. Or start an OpenAI-compatible server at http://localhost:8080/v1
    mlx_lm.server --model lmstudio-community/Qwen2.5-3B-Instruct-MLX-4bit

Official weights and documentation: huggingface.co/Qwen ↗

$ how the score was worked out · microsoft/Phi-4-mini-instruct… 52%, WildEval ZeroEval published r… 34%, SpeakLeash CPTUB (Complex Pol… 3% and 8 more

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

sourceresultsshare of the scorelicence
microsoft/Phi-4-mini-instruct model card16▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 52%MIT
WildEval ZeroEval published results4▇▇▇▇▇▇▇▇▇▇ 34%Apache-2.0
SpeakLeash CPTUB (Complex Polish Text Understanding Benchmark)1▇ 3%MIT
Inception Arabic/Hindi leaderboards (AraGen-v3, HindiGen-v1, Arabic IFEval)1▇ 3%Apache-2.0
CodeReviewQA: The Code Review Comprehension Assessment for Large Language Models (arXiv:2503.16167v2)4▇ 2%CC-BY-4.0
SpeakLeash Polish EQ-Bench leaderboard1▇ 2%Apache-2.0
BRIDGE Medical Leaderboard (Harvard Medical School / Brigham and Women's YLab)1▇ 2%MIT
Open Persian LLM Leaderboard v2 (Part DP AI and AUT NLP Lab)1▇ 2%Apache-2.0
Open Portuguese LLM Leaderboard (archived)1▇ 1%Apache-2.0
Berkeley Function Calling Leaderboard result archive1▇ 1%Apache-2.0
Vectara hallucination leaderboard (vectara/hallucination-leaderboard repository)1▇ 0%Apache-2.0
prior · pull toward the middle▇ 0%
weighted average= 42.7

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated · sources and licences · json