[whatmodel.sh](../../README.md) / Qwen3 0.6B
# Qwen3 0.6B
Alibaba · released 2025-04 · 0.75B params · apache-2.0
**score 35.9** ±0.8 · #619 of 732 · from 27 benchmark results across 11 sources
Score within the [LMArena (text)](https://whatmodel.sh/?set=lmarena-text&score=set) comparison set only.
## What it's good at
● marks Qwen3 0.6B among every model on that task or test, worst to best.
task vs the others rank counts tests
agent tasks ▁▁▃●█▄▅▅▄▄▂▁ #279 of 313* 3.3% 2
coding ▁●▁▃▅▇█▇▅▄▃▁ #461 of 469* 6.9% 2
chat & writing ▁▁▁▃▆●▆█▆▇▇▃ #308 of 402* 3.4% 1
math ▁▁▃▅●▆▄█▄▂▁▁ #325 of 431 35% 4
agent tasks · #279 of 313 · 3.3% of its score
Beats [Magistral Small 1.0 (2506)](magistral-small-2506.md), [Llama 4 Scout (109B-A17B)](llama-4-scout.md), [Bielik 11B v2.3 Instruct](bielik-11b-v2-3-instruct.md) here, though they rank higher overall.
test result vs the others in its run difficulty counts why
BFCL V4 (December 2025 archive) 23.9% ▁▁▁▂●█▅▅▆▂▂▂ beats 12/77 hard · avg 31% 1.8% typical
BFCL v3 47.3% ▁▁▁▁▁▃●█▅▄▁▁ beats 29/114 medium · avg 51% 1.5% blunt · at its level
coding · #461 of 469 · 6.9% of its score
Beats [Molmo 7B-D (0924)](molmo-7b-d.md), [DeepSeek R1 Distill Qwen 1.5B](deepseek-r1-distill-qwen-1-5b.md) here, though they rank higher overall. Trails [Phi-1 (1.3B)](phi-1.md), [Pythia 6.9B](pythia-6-9b.md), [GPT-J 6B](gpt-j-6b.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
HumanEval+ 14.8% ●·······▂·▂█ beats 0/6 easy · avg 78% 4.2% sharp · at its level · noisy
LiveCodeBench v3 34.4% ▃●·····▃·▃·█ beats 1/6 medium · avg 64% 2.7% at its level
chat & writing · #308 of 402 · 3.4% of its score
Beats [Ministral 3 14B Reasoning 2512](ministral-3-14b-reasoning.md), [Gemma 4 E4B](gemma-4-e4b.md), [OpenThinker3 7B](openthinker3-7b.md) here, though they rank higher overall. Trails [LFM2 350M](lfm2-350m.md), [Gemma 3 1B](gemma-3-1b.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
IFEval (aggregation unspecified) 62.3% ▄●·····█▄▄·▄ beats 1/6 easy · avg 82% 3.4% at its level · several runs
math · #325 of 431 · 35% of its score
Beats [GPT-4 (0613)](gpt-4-0613.md), [Athene-70B](athene-70b.md), [Gemma 2 27B IT](gemma-2-27b.md) here, though they rank higher overall. Trails [Stable Code 3B](stable-code-3b.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
GSM8K 80.9% ●··▃··▃▃·▃·█ beats 0/7 very easy · avg 93% 12% sharp · at its level · noisy · several …
MGSM 41.3% █···●▄··▄··▄ beats 2/6 easy · avg 83% 11% sharp · at its level
AIME 2024 11.2% ●·····▄··▄██ beats 0/6 medium · avg 58% 6.5% sharp
AIME 2025 15.2% ●··▄····█▄▄▄ beats 0/6 medium · avg 50% 5.7% sharp
other tests (14)
test result vs the others in its run difficulty counts why
MMLU 55.8% ●···▄··█··▄█ beats 0/6 easy · avg 76% 15% at its level · several runs
BIG-Bench Hard 41.5% ●·····▄·▄·██ beats 0/6 easy · avg 77% 10% sharp · at its level
GPQA (subset unspecified) 27.2% ●······▃·▅·█ beats 0/6 hard · avg 43% 4.3% sharp · too hard for it · several runs
Open Persian LLM Leaderboard v2 a… 36.3% ▂●▂▃█▄▃▆▃▅▂▂ beats 7/81 medium · avg 57% 3.2% blunt · at its level
IFBench (aggregation unspecified) 19.8% ▄▄·▄····●·█▄ beats 3/6 medium · avg 38% 3.0% typical
BRIDGE clinical text, zero-shot a… 20.4% ▁▁▁▁▁●▂▃▄▆█▃ beats 11/97 hard · avg 31% 2.9% blunt
MMMLU 30.8% ●···▄·▄▄··▄█ beats 0/6 medium · avg 57% 2.9% typical
ZebraLogic 29.8% ●▄····▄·▄·▄▄ beats 0/6 medium · avg 52% 2.3% typical
Open PT-LLM Leaderboard average (… 50.3% ▃▁▁▁▁▁▁●▂█▅▁ beats 68/276 medium · avg 72% 1.5% at its level · noisy
KINA (Knowledge Index of Noah's A… 14.5% ●▁▂▁▃▄█▆▄▅▂▃ beats 0/39 very hard · avg 21% 1.4% too hard for it
AraGen-v3 3C3H score (Inception, … 1.6% ▁●·▁▂▂▄█▄▅▄▄ beats 1/66 hard · avg 29% 1.4% typical
EuroEval Life in the UK (English … 61.9% ▁▁▁▁▁▁▁●▄█▄▃ beats 32/146 easy · avg 82% 1.2% at its level · noisy
EuroEval HellaSwag (English), tes… 38.6% ▃·▁▁▂●▂▅█▇▄▁ beats 19/100 medium · avg 67% 0.9% sharp · noisy
Vectara hallucination leaderboard… 96.3% ▁▁▂▃▄▄█●▃▂▂▁ beats 91/140 very easy · avg 94% 0.8% blunt · noisy
source results share of the score
Olmo 3 9 42% ▇▇▇▇▇
LiquidAI/LFM2-1.2B model card 7 33% ▇▇▇▇
Unsupervised Confidence Calibration for… 1 8% ▇
Berkeley Function Calling Leaderboard r… 2 3% ▇
Open Persian LLM Leaderboard v2 1 3% ▇
BRIDGE Medical Leaderboard 1 3% ▇
EuroEval leaderboards 2 2% ▇
Open Portuguese LLM Leaderboard 1 1% ▇
Knowledge Index of Noah's Ark 1 1% ▇
Inception Arabic/Hindi leaderboards 1 1% ▇
Vectara hallucination leaderboard 1 1% ▇
prior: pull toward the middle 0% ▇
weighted average = 35.9
Licences: Olmo 3 ([CC BY 4.0](../../licenses/report-2512.13961v2.txt)) · LiquidAI/LFM2-1.2B model card ([LFM Open License v1.0](https://huggingface.co/LiquidAI/LFM2-1.2B/blob/40f3da0d0164913923aee9462c23077868b816a3/LICENSE)) · Unsupervised Confidence Calibration for Reasoning LLMs from a Single Generation ([CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)) · Berkeley Function Calling Leaderboard result archive ([Apache-2.0](../../licenses/bfcl-result-archive-v4-Apache-2.0.txt)) · Open Persian LLM Leaderboard v2 ([Apache-2.0](../../licenses/hf-leaderboard-spaces-Apache-2.0.txt)) · BRIDGE Medical Leaderboard ([MIT](../../licenses/bridge-medical-leaderboard-MIT.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)) · Knowledge Index of Noah's Ark ([CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)) · Inception Arabic/Hindi leaderboards ([Apache-2.0](../../licenses/hf-leaderboard-spaces-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 Qwen3 0.6B (Q4_K_M, needs ~0.5 GB, from [huggingface.co/lmstudio-community](https://huggingface.co/lmstudio-community/Qwen3-0.6B-GGUF/tree/3334d820ab76652cf6e242d7c6302b10f0951f23))
```sh
llama cli -hf lmstudio-community/Qwen3-0.6B-GGUF:Q4_K_M
```
3. Or start a local server and open the address it prints (http://localhost:9931)
```sh
llama serve -hf lmstudio-community/Qwen3-0.6B-GGUF:Q4_K_M
```
Other versions: Q8_0 (~0.8 GB) `llama cli -hf lmstudio-community/Qwen3-0.6B-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 0.6B (Q4_K_M, needs ~0.5 GB, from [huggingface.co/lmstudio-community](https://huggingface.co/lmstudio-community/Qwen3-0.6B-GGUF/tree/3334d820ab76652cf6e242d7c6302b10f0951f23))
```sh
lms get https://huggingface.co/lmstudio-community/Qwen3-0.6B-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 (~0.8 GB) `lms get https://huggingface.co/lmstudio-community/Qwen3-0.6B-GGUF@Q8_0 -y` · 4bit (~0.3 GB, Mac only) `lms get https://huggingface.co/lmstudio-community/Qwen3-0.6B-MLX-4bit -y` · 8bit (~0.6 GB, Mac only) `lms get https://huggingface.co/mlx-community/Qwen3-0.6B-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 0.6B (Q4_K_M, needs ~0.95 GB, from [ollama.com](https://ollama.com/library/qwen3/tags))
```sh
ollama run qwen3:0.6b
```
Other versions: Q8_0 (~1.4 GB) `ollama run qwen3:0.6b-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 0.6B (4bit, needs ~0.3 GB, Mac only, from [huggingface.co/lmstudio-community](https://huggingface.co/lmstudio-community/Qwen3-0.6B-MLX-4bit/tree/75429955681c1850a9c8723767fe4252da06eb57))
```sh
mlx_lm.chat --model lmstudio-community/Qwen3-0.6B-MLX-4bit
```
3. Or start an OpenAI-compatible server at http://localhost:8080/v1
```sh
mlx_lm.server --model lmstudio-community/Qwen3-0.6B-MLX-4bit
```
Other versions: 8bit (~0.6 GB, Mac only) `mlx_lm.chat --model mlx-community/Qwen3-0.6B-8bit`
Model repository and documentation: [huggingface.co/Qwen](https://huggingface.co/Qwen/Qwen3-0.6B)
## 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… Q8_0 208.73 tok/s llama.cpp 32e789fdf (build_number…
NVIDIA DGX Spark (GB10) BF16 119.4 tok/s llama.cpp (CUDA build, GGUF BF16 …
NVIDIA DGX Spark (GB10) BF16 103.3 tok/s vLLM (CUDA, Safetensors BF16; ver…
NVIDIA DGX Spark (GB10) BF16 106.8 tok/s SGLang (CUDA, Safetensors BF16; v…
NVIDIA DGX Spark (GB10) BF16 55.3 tok/s llama.cpp (CUDA build, GGUF BF16 …
NVIDIA DGX Spark (GB10) BF16 80.2 tok/s vLLM (CUDA, Safetensors BF16; ver…
NVIDIA DGX Spark (GB10) BF16 82.6 tok/s SGLang (CUDA, Safetensors BF16; v…
Apple M4 Pro (16-core GPU, … Q4 GGUF (exact … 297.4 tok/s llama.cpp b9630 (June 2026), Meta…
Apple M4 Pro (16-core GPU, … Q4 MLX (4-bit; … 343.6 tok/s MLX v0.31.2 via mlx-lm, default g…
Apple M4 Pro (16-core GPU, … Q8 GGUF (exact … 219.8 tok/s llama.cpp b9630 (June 2026), Meta…
Apple M4 Pro (16-core GPU, … Q8 MLX (8-bit; … 255.3 tok/s MLX v0.31.2 via mlx-lm, default g…
Apple M3 (8-core GPU, 16GB) Q4 GGUF (exact … 180.5 tok/s llama.cpp b9630 (June 2026), Meta…
Apple M3 (8-core GPU, 16GB) Q4 MLX (4-bit; … 182.3 tok/s MLX v0.31.2 via mlx-lm, default g…
Apple M3 (8-core GPU, 16GB) Q8 GGUF (exact … 99.8 tok/s llama.cpp b9630 (June 2026), Meta…
Apple M3 (8-core GPU, 16GB) Q8 MLX (8-bit; … 113.6 tok/s MLX v0.31.2 via mlx-lm, default g…
Apple M5 Pro (48GB; CPU/GPU… Q4 GGUF (4-bit;… 386.1 tok/s llama.cpp b9960, Metal backend, d…
Apple M5 Pro (48GB; CPU/GPU… MLX 4-bit (grou… 398.1 tok/s mlx-lm 0.31.3 (MLX 0.32.0), defau…
Apple M5 Pro (48GB; CPU/GPU… Q8 GGUF (8-bit;… 285.3 tok/s llama.cpp b9960, Metal backend, d…
Apple M5 Pro (48GB; CPU/GPU… MLX 8-bit (grou… 295.5 tok/s mlx-lm 0.31.3 (MLX 0.32.0), defau…
Apple M2 Pro (16-core GPU),… Q4_K_M 214.58 tok/s llama.cpp · whatmodel bench.py 3.…
Apple M2 Pro (16-core GPU),… Q4_K_M 167.16 tok/s lm-studio · whatmodel bench.py 3.…
Radeon 840M / 860M Graphics… q4_k_m · layers… 61.36 tok/s ollama (version not recorded) · l…
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 dynamic (weights FP8 per-… OpenLLM v1 average of 6 tasks (MM… 0.378 → 0.378 100.1%
FP8 dynamic (weights FP8 per-… MMLU — published row 'MMLU (5-sho… 0.428 → 0.423 98.8%
FP8 dynamic (weights FP8 per-… GPQA Diamond — published row 'GPQ… 0.293 → 0.258 87.9%
GPTQ INT8 (w8, group_size 128… GPQA (subset not stated in card),… 0.279 → 0.293 105.0%
GPTQ INT8 (w8, group_size 128… GPQA (subset not stated in card),… 0.229 → 0.232 101.3%
GPTQ INT8 (w8, group_size 128… MMLU-Redux, non-thinking mode (Qw… 0.446 → 0.449 100.7%
GPTQ INT8 (w8, group_size 128… LiveBench 2024-11-25, non-thinkin… 0.218 → 0.218 100.0%
GPTQ INT8 (w8, group_size 128… LiveBench 2024-11-25, thinking mo… 0.303 → 0.302 99.7%
GPTQ INT8 (w8, group_size 128… MMLU-Redux, thinking mode (Qwen-r… 0.556 → 0.538 96.8%
INT4 W4A16 (weight-only INT4,… OpenLLM v1 average of 6 tasks (MM… 0.378 → 0.362 95.8%
INT4 W4A16 (weight-only INT4,… GPQA Diamond — published row 'GPQ… 0.293 → 0.278 94.8%
INT4 W4A16 (weight-only INT4,… MMLU — published row 'MMLU (5-sho… 0.428 → 0.398 92.9%
Each kept value links to its source.
[interactive version](https://whatmodel.sh/model/qwen3-0-6b) · [JSON](https://whatmodel.sh/api?format=json&all=1)