Qwen3.6 27B

+ contribute

Qwen3.6 27B is an open-weights language model from Alibaba, released April 2026. Its published size is 27.8B parameters and a 262k-token context window. Licence: apache-2.0.

69.4±0.6#91 of 719 LLMsfrom 79 benchmark results across 10 sources

$ how it compares

The LLMs ranked either side of Qwen3.6 27B, out of 828 with a score, then more from Alibaba.

#	model	score	where	RAM	released
97	Muse Glimmer 30B Meta	70±1	api $0.64/M +local	~19GB	2026-08
98	Nemotron 3 Ultra 550B-A55B NVIDIA	69±1	api $1.35/M +local	~89GB	2026-06
99	Qwen3.6 27B Alibaba	69±1	api $1.20/M +local	~18GB	2026-04
100	Ring-2.6-1T inclusionAI (Ant Group)	69±2	comparison only		2026-05
101	A.X K2 (688B-A33B) SK Telecom	69±1	local	~1454GB	2026-07
	more from Alibaba
18	Qwen3.8 Max (2.4T-A95B) Alibaba	78±1	api $3.00/M +local	~380GB	2026-08
34	Qwen3.8 Max (0902) Alibaba	76±1	api $3.00/M		2026-09
37	Qwen3.8-Flash-Next Alibaba	75±1	local	~59GB	2026-08

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #73 of 300 · 20% of its score · 12 tests

Beats GPT-5 Pro, GPT-5, Gemini 3 Flash (Preview) here, though they rank higher overall. Trails MiMo-V2.5 (310B-A15B) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
SWE-bench Pro public set, Qwen-refined lab +1	53.5% ✎	▄······▄█·▄▄ beats 4/5	hard · avg 18%	▇ 3.0%	at its level · several runs
SWE-bench Verified lab	77.2% ✎	▁▁▁▁▁▁▂▄█▆▁▁ beats 112/131	hard · avg 26%	▇ 2.8%	sharp
SWE-bench Pro (split unspecified)	50.2% ✎	▃·▁▁▁▂▂▃█▇▃▁ beats 26/55	very hard · avg 8.3%	▇ 2.4%	sharp · at its level · small run
OSWorld-Verified lab +1	63.9% ✎	█▄·▂▆▂▄▂▆▄▆█ beats 16/25	hard · avg 26%	▇ 2.4%	at its level · small run · several runs
Terminal-Bench 2.1 lab	63.4% ✎	▁·▁·▁▁▄▃▄█▃▇ beats 24/56	hard · avg 16%	▇ 2.1%	at its level · small run
Terminal-Bench 2.0 lab	59.3% ✎	▂▂▅▅████▆▂▃▃ beats 26/36	very hard · avg 14%	▇ 1.5%	at its level · small run
SWE-bench Multilingual lab	71.3% ✎	▁··▁▃▂▅▇█▃▃▁ beats 35/60	hard · avg 29%	▇ 1.4%	at its level · small run
Thinkingbox-bench pass@1 (507 stateful business-workflow agent tasks, N=20)	32.9% ✎	▅··▃·▃·█▃▅██ beats 5/15	very hard · avg 4.9%	▇ 1.1%	at its level

4 more tests

test	result	vs the others in its run	difficulty	counts	why
MCP-Atlas (public set)	62.5% ✎	▂·▁▂▁·▁▃█▆▆▂ beats 13/37	hard · avg 28%	▇ 0.9%	at its level · small run
DeepSearchQA	71.1% ✎	█·······█··█ beats 1/2	medium · avg 40%	▇ 0.8%	at its level · small run
EEGAgentBench overall (tool-based EEG analysis agents, 6 tasks)	43.3% ✎	▂·▁▁··▇▇█▂·▁ beats 6/25	hard · avg 29%	▇ 0.8%	blunt · at its level
WebArena-Verified lab	48.8% ✎	█···█······█ beats 0/2	hard · avg 22%	▇ 0.7%	at its level · small run

coding▁▁▁▃▆██▇▆▄▃▁ #78 of 458 · 15% of its score · 7 tests

Beats GPT-5.2, Grok 4.20 (0309 reasoning), GPT-5.4 mini here, though they rank higher overall. Trails Llama 3.1 Nemotron Ultra 253B v1, GLM-4.7, Gemini 3.5 Flash-Lite here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
SWE-bench Pro public set, Qwen-refined lab +1	53.5% ✎	▄······▄█·▄▄ beats 4/5	hard · avg 18%	▇ 3.0%	at its level · several runs
SWE-bench Verified lab	77.2% ✎	▁▁▁▁▁▁▂▄█▆▁▁ beats 112/131	hard · avg 26%	▇ 2.8%	sharp
SWE-bench Pro (split unspecified)	50.2% ✎	▃·▁▁▁▂▂▃█▇▃▁ beats 26/55	very hard · avg 8.3%	▇ 2.4%	sharp · at its level · small run
Luau Leetcode fully solved (Torpedo Software)	51.9% ✎	▁▁▁▁·▁▃▃▅▃█▇ beats 9/49	hard · avg 18%	▇ 2.3%	at its level
LiveCodeBench v6 (window unspecified) lab	83.9% ✎	▂▃▄▄▅██▅▄▇▃▆ beats 49/65	medium · avg 46%	▇ 1.8%	small run
LiveCodeBench v6 (Feb–May 2025) lab	83.9% ✎	▁▁▃▁█▅▄▇▅▅▃█ beats 34/38	medium · avg 47%	▇ 1.5%	typical
SWE-bench Multilingual lab	71.3% ✎	▁··▁▃▂▅▇█▃▃▁ beats 35/60	hard · avg 29%	▇ 1.4%	at its level · small run

math▁▂▃▅▆▅▄█▄▁▁▁ #79 of 398 · 25% of its score · 9 tests

Beats Claude Sonnet 4.6, Claude Opus 4.5, Qwen3.5 397B-A17B here, though they rank higher overall. Trails Granite 4.2 30B, INTELLECT-3, EXAONE 4.5 33B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
FrontierMath tiers 1–3 v2	35.1% ✎	▂··▁▁▃▄▄█▅▃▃ beats 25/82	very hard · avg 1.2%	▇▇▇▇▇ 12%	sharp · at its level
HMMT February 2026 lab	84.3% ✎	▃▃▂▃·██▆█▅▃▅ beats 24/33	medium · avg 36%	▇ 3.2%	sharp
AIME 2026 lab	94.1% ✎	▁▁▁▂▁▂▆█▆▂·▁ beats 36/49	medium · avg 58%	▇ 2.5%	sharp · maxed out · small run
IMO-AnswerBench lab	80.8% ✎	▁▁·▁▁▁▄▅█▄▂▃ beats 23/42	medium · avg 36%	▇ 1.9%	small run
OTIS mock AIME 2024–2025	91.1% ✎	▃▃▄▃▃▃▇▇█▃▂▅ beats 147/193	very hard · avg 8.5%	▇ 1.4%	sharp · noisy
HMMT February 2025 lab	93.8% ✎	▁▁▁▃▁▅▅▃██▁▄ beats 23/31	medium · avg 62%	▇ 1.3%	maxed out · small run
MathVista (testmini) lab	87.4% ✎	▁··▁▄▄▇█▅▆█▄ beats 44/47	medium · avg 63%	▇ 0.9%	small run
HMMT November 2025 lab	90.7% ✎	▁▁····▁▅▃█▇▁ beats 15/23	easy · avg 66%	▇ 0.9%	small run

1 more test

test	result	vs the others in its run	difficulty	counts	why
DynaMath lab	85.6% ✎	▂·····▂··█▄▆ beats 7/10	medium · avg 62%	▇ 0.7%	small run

vision & documents▁▁▃▂▁▃▄▂▆▇█▄ #48 of 147 · 10% of its score · 14 tests

Beats MiMo-V2.6-Pro, DeepSeek V4 Flash Vision (Exp), GPT-5 here, though they rank higher overall. Trails Qwen3.6 Flash, Gemma 4 31B, Gemini 3.5 Flash-Lite here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
CharXiv (reasoning questions) lab +1	78.4% ✎	▂▅▃▅▅▆▆▅▆█▆▂ beats 73/100	medium · avg 43%	▇ 2.3%	small run · several runs
OmniDocBench 1.5 overall (higher is better) lab +1	89.4% ✎	▂··▂▂▂··▂▂█▅ beats 8/13	easy · avg 69%	▇ 1.8%	small run · several runs
MathVista (testmini) lab	87.4% ✎	▁··▁▄▄▇█▅▆█▄ beats 44/47	medium · avg 63%	▇ 0.9%	small run
MMMU (val) lab	82.9% ✎	▁····▁▁▄█▇▆▆ beats 50/52	medium · avg 57%	▇ 0.9%	small run
MMMU Pro lab	75.8% ✎	▂▃▁▃▆▇▃▇▇██▅ beats 41/52	medium · avg 46%	▇ 0.8%	at its level · small run
MMStar lab	81.4% ✎	▃▅▂▃▂▂▆█▆▆█▅ beats 31/34	medium · avg 62%	▇ 0.6%	small run
CountBench lab	97.8% ✎	▂▂▃▂·▂▂▃▃·▅█ beats 16/18	very easy · avg 90%	▇ 0.5%	maxed out · small run
RealWorldQA lab +1	84.1% ✎	▂▃▂█▃██▆█▂▆▆ beats 33/38	easy · avg 72%	▇ 0.5%	blunt · small run · several runs

6 more tests

test	result	vs the others in its run	difficulty	counts	why
SimpleVQA lab	56.1% ✎	▃·▃·▃·▃·█▅▃▃ beats 5/10	hard · avg 34%	▇ 0.5%	at its level · small run
ScreenSpot-Pro	76.1% ✎	▂▃▅▆██▅▂···▂ beats 22/24	medium · avg 51%	▇ 0.5%	blunt · at its level · small run
MMBench-EN dev v1.1 lab	92.3% ✎	█▄···▂··▄▂▄▆ beats 10/14	easy · avg 82%	▇ 0.4%	blunt · maxed out · small run
CC-OCR lab	81.2% ✎	▁·▂··▂▂▃▃▂▆█ beats 21/26	easy · avg 72%	▇ 0.2%	blunt · small run
OCRBench lab	89.4% ✎	▁·▃▁▆▅▂█▅█▁▃ beats 28/37	easy · avg 83%	▇ 0.2%	blunt · small run
PureDocBench document parsing (Avg3 over Clean/Digital/Real)	65.6% ✎	▂·▁·▂▃▅█·▁▂▂ beats 5/22	medium · avg 62%	▇ 0.1%	blunt · at its level

web search▂▁▂▆▄█▃▃▁▁·▁ #49 of 80 * · 0.8% of its score · 1 test

Beats Gemini 3.1 Pro (Preview), GPT-5.2, Kimi K2.6 here, though they rank higher overall. Trails GPT-4o mini (2024-07-18), Mistral Small 3.1 24B (2503), Llama 4 Scout (109B-A17B) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
DeepSearchQA	71.1% ✎	█·······█··█ beats 1/2	medium · avg 40%	▇ 0.8%	at its level · small run

other tests30 not in a task · 40% of its score

test	result	vs the others in its run	difficulty	counts	why
GPQA Diamond +2	85.9% ✎	▁▁▁▁▂▄▆▄▅▇█▅ beats 150/208	hard · avg 45%	▇▇▇▇ 11%	sharp · several runs
Humanity's Last Exam lab +1	24.0% ✎	▁▁▁▃▆▆▇██▆▁▃ beats 54/104	very hard · avg 5.1%	▇ 3.4%	several runs
NL2Repo lab +1	36.2% ✎	▄·····█·▄▄·▄ beats 4/5	very hard · avg 8.2%	▇ 2.7%	at its level · several runs
MMLU-Pro lab	86.2% ✎	▁▁·▁▁▁▃▂▄▅█▁ beats 180/188	medium · avg 56%	▇ 1.8%	typical
IFBench (aggregation unspecified) lab +1	69.1% ✎	▃▇▂▄▅▅▂▄▄▆█▇ beats 56/80	medium · avg 40%	▇ 1.7%	at its level · small run · several runs
Mystery Game Puzzles (Epoch)	7.0% ✎	█··▂▃▇▄█▂▃▁▁ beats 10/74	very hard · avg 9.4%	▇ 1.6%	sharp · too hard for it · noisy
Chess Puzzles (Epoch)	22.0% ✎	█▁▁▁▁▂▁▄▂▂▁▁ beats 90/135	very hard · avg 5.9%	▇ 1.4%	sharp · noisy
ERQA lab +1	62.5% ✎	▃··▃··█·▃█▅▅ beats 8/12	hard · avg 35%	▇ 1.3%	at its level · small run · several runs

22 more tests

test	result	vs the others in its run	difficulty	counts	why
AndroidWorld lab +1	70.3% ✎	▁·▁▁▂▃█▅▂▃·▂ beats 21/24	medium · avg 46%	▇ 1.1%	blunt · at its level · small run · several runs
C-Eval lab	91.4% ✎	▁···▁▁▂▁▂▂▃█ beats 28/33	medium · avg 69%	▇ 1.1%	small run
WildClawBench	43.2% ✎	▂·▄▄▆▂▄█▆▂▂▂ beats 13/20	very hard · avg 14%	▇ 1.1%	at its level · small run
QwenClawBench (Qwen internal) lab	53.4% ✎	█·█··█·····█ beats 3/3	hard · avg 23%	▇ 1.0%	at its level
MMLU-Redux lab	93.5% ✎	▁▁▃▁▂▃▄▅▇▅█▄ beats 57/63	easy · avg 78%	▇ 1.0%	maxed out · small run
CPTUB average (SpeakLeash, 0-5 judge scale)	88.2% ✎	▁▂▂▂▃█▃▆▅▄▂▃ beats 94/95	easy · avg 66%	▇ 0.9%	typical
RobloxQA v2.0 multiple choice (Torpedo Software)	89.9% ✎	▁▁▃▁█▄▃▃▆▂▂▂ beats 15/49	easy · avg 78%	▇ 0.9%	typical
Claw-Eval Pass^3 lab	60.6% ✎	█▄▄·▄·▄▄·▄·▄ beats 6/8	hard · avg 24%	▇ 0.8%	at its level
Gaia2	40.0% ✎	█·····█····█ beats 1/2	hard · avg 16%	▇ 0.7%	at its level · small run
SkillsBench 78-task subset (OpenCode, avg of 5) lab	48.2% ✎	█▄·········▄ beats 3/3	very hard · avg 9.9%	▇ 0.7%	at its level
SkillsBench (with skills)	46.6% ✎	█·········██ beats 2/2	hard · avg 16%	▇ 0.7%	at its level · small run
SuperGPQA lab	66.0% ✎	▂▁▂▂▅▆▃▇▇▇█▃ beats 43/46	medium · avg 36%	▇ 0.7%	at its level · small run
VideoMMMU lab	84.4% ✎	▂···▂···▂▂█▆ beats 9/10	medium · avg 61%	▇ 0.7%	small run
VLMs Are Blind lab	97.0% ✎	▂·▂·····▃··█ beats 7/8	very easy · avg 88%	▇ 0.6%	maxed out · small run
Video-MME (with subtitles) lab	87.7% ✎	▁▁▁▁▁·▁▃▁▁▁█ beats 16/16	medium · avg 71%	▇ 0.6%	small run
RefSpatial-Bench lab	70.0% ✎	▃·▃····▃▃▅▅█ beats 9/10	medium · avg 40%	▇ 0.6%	at its level · small run
MLVU lab	86.6% ✎	▂···▂···▂▄▄█ beats 8/10	easy · avg 69%	▇ 0.6%	small run
Claw-Eval average score lab	72.4% ✎	▄·▄··█▄▄▄▄▄▄ beats 9/9	medium · avg 47%	▇ 0.5%	blunt · at its level
EmbSpatial-Bench lab	84.6% ✎	▁·····▁·▁·▃█ beats 10/10	easy · avg 67%	▇ 0.5%	small run
WMT25 Terminology Translation (term success rate)	89.2% ✎	▂·▂··▂·▄▂·█▆ beats 6/12	medium · avg 64%	▇ 0.4%	noisy
MVBench lab	75.5% ✎	▁··▃▄▁▃▄▄▄█▄ beats 23/26	medium · avg 66%	▇ 0.2%	blunt · at its level · small run
WMT26 General MT (GEMBA-ESA, LLM judge)	61.9% ✎	▄·▄▄▄·▄██▄█▄ beats 0/12	easy · avg 65%	▇ 0.2%	blunt

$ where to run it

Through an API

api · Groq · $1.20/M · $0.6 in / $3 out per 1M tokens · 262k context · console.groq.com ↗ · price sourceStandard base token rate from models.dev. Provider-specific tiers, long-context surcharges, caching and promotions may differ.

llama.cpp · recommended

version

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 ↗
Run Qwen3.6 27B · needs ~19.1 GB · from huggingface.co/ggml-org
chat in the terminal
llama cli -hf ggml-org/Qwen3.6-27B-GGUF:Q4_K_M
Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf ggml-org/Qwen3.6-27B-GGUF:Q4_K_M

LM Studio

version

Install LM Studio · Open the app once before using the lms command.
download it (macOS / Windows / Linux) ↗
Run Qwen3.6 27B · needs ~19.1 GB · from huggingface.co/ggml-org
open in LM Studio ↗
Or download it in the terminal
lms get https://huggingface.co/ggml-org/Qwen3.6-27B-GGUF@Q4_K_M -y
Then run lms ls to see its name, and lms chat followed by that name

Ollama

version

Install Ollama
download it (macOS / Windows) ↗
Linux
curl -fsSL https://ollama.com/install.sh | sh
Run Qwen3.6 27B · needs ~17.18 GB · from ollama.com
chat in the terminal
ollama run qwen3.6:27b

MLX (Mac)

version

Install MLX (Mac) · Apple silicon Macs only, text only.
with uv
uv tool install mlx-lm
or
pip install mlx-lm
Run Qwen3.6 27B · needs ~16.1 GB, Mac only · from huggingface.co/lmstudio-community
chat in the terminal
mlx_lm.chat --model lmstudio-community/Qwen3.6-27B-MLX-4bit
Or start an OpenAI-compatible server at http://localhost:8080/v1
mlx_lm.server --model lmstudio-community/Qwen3.6-27B-MLX-4bit

Official weights and documentation: huggingface.co/Qwen ↗

$ how the score was worked out · Qwen/Qwen3.6-27B model card 45%, Epoch AI — internal evaluatio… 20%, Qwen/Qwen3.8-27B model card 18% and 7 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

source	results	share of the score	licence
Qwen/Qwen3.6-27B model card · lab-reported	42	▇▇▇▇▇▇▇▇▇▇▇▇▇ 45%	apache-2.0
Epoch AI — internal evaluations	5	▇▇▇▇▇▇ 20%	CC BY 4.0
Qwen/Qwen3.8-27B model card · lab-reported	14	▇▇▇▇▇ 18%	apache-2.0
meta-models/Muse-Glimmer-30B model card (speed and benchmark tables)	10	▇▇▇ 10%	Apache-2.0
Roblox LLM Leaderboard (Torpedo Software): RobloxQA v2.0 and Luau Leetcode	2	▇ 3%	MIT
One Success Isn't Reliability: Thinkingbox, a Sandbox and Benchmark for Agents in Stateful Business Workflows (arXiv:2608.19741v4)	1	▇ 1%	CC BY 4.0
SpeakLeash CPTUB (Complex Polish Text Understanding Benchmark)	1	▇ 1%	MIT
EEGAgentBench: Benchmarking LLM Agents on Short- and Long-Horizon EEG Analysis (arXiv:2609.31632v1)	1	▇ 1%	CC BY 4.0
Cohere — North Small Translate (arXiv:2609.13916v1)	2	▇ 1%	CC BY 4.0
How Far Is Document Parsing from Solved? PureDocBench: A Source-Traceable Benchmark across Clean, Degraded, and Real-World Settings (arXiv:2605.07492v2)	1	▇ 0%	CC BY 4.0
prior · pull toward the middle		▇ 0%
weighted average		= 69.4

$ published speed measurements · 2

Reference results for the hardware, engine and settings below. They aren't verified speeds of the downloads above or predictions for your machine.

Apple M5 Pro (48GB; CPU/GPU core counts not stated)	Q4 GGUF (4-bit; exact llama.cpp type not stated)	15.4 tok/s	llama.cpp b9960, Metal backend, default settings · tg128: token-generation throughput over 128 generated tokens; average of five repetitions; macOS (Darwin 25.4) Baseline runtime column from the BaseRT M5 paper; BaseRT's own column is not recorded.	source ↗
Apple M5 Pro (48GB; CPU/GPU core counts not stated)	MLX 4-bit (group size not stated)	18.4 tok/s	mlx-lm 0.31.3 (MLX 0.32.0), default generation settings · tg128: token-generation throughput over 128 generated tokens; average of five repetitions; macOS (Darwin 25.4) Baseline runtime column from the BaseRT M5 paper; BaseRT's own column is not recorded.	source ↗

$ improve this page

Update model information Add a place to run it Add a test result Propose a benchmark Correct a benchmark score Edit 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 2026-10-11 02:20 UTC · sources and licences · json