LOADING DATASHEET
GROUNDED IN REAL HUMAN CONVERSATIONSYOUR BENCH
Qwen3.8 27B
◆ OPEN
by Alibaba
OPEN WEIGHTSLOCALCODING HELPWRITING HELPFAST ANSWERS
RANKED #2 OF 316 ASSISTANTS · OVERALL #2 OF 493 RANKED · VIBE SCORE 7.6 · 21,155 VOICES ·
COPY BADGE pastes a README snippet with the rank SVG.
A fast, open text model from Alibaba that runs well on local hardware.
HAPPY9.7
2.9k mentions
SAD5.3
744 mentions
VIBE TRAJECTORY
26 weeks · 21,733 voices in this window
QUIETLY DEGRADING
Recent vibe and chatter are both sliding.
HOW TO USE IT
Open weights. You can use a hosted service, or download it and run it yourself, free.
RUN IT ON YOUR OWN MACHINE
Open weights: download Qwen3.8 27B once and it's yours. No subscription, no rate limits, works offline.
27.8B parameters, a Q4_K_M file is about 17 GB
- BARE MINIMUM
Mac Mini M2 (24GB Unified Memory)
24 GB UNIFIED MEMORY
~$650 USEDQ4_K_M, ~12 tok/s, 16k context
- SWEET SPOT
Nvidia GeForce RTX 3090 24GB
24 GB VRAM
~$1,100 USEDQ4_K_M, ~35 tok/s, 32k context
- FULL POWER
Mac Studio M2 Ultra (128GB Unified Memory)
128 GB UNIFIED MEMORY
~$4,200 USEDBF16 unquantized, ~30 tok/s, full 262k context
LEARN HOW TO SET IT UP
WILL MY BOX RUN QWEN3.8 27B?
The three cards above are researched picks. Search the machine you actually have — we only say yes if it should reply at a usable speed, not just load the weights and crawl.
HARDWARE PICKS & PRICES ARE RESEARCHED FROM THE LIVE WEB AND REFRESHED AUTOMATICALLY. TREAT THEM AS BALLPARK, NOT GOSPEL.
WHAT CAN YOU ACTUALLY DO WITH IT?
Everyday jobs on each of those rigs: how long Qwen3.8 27B takes, and how good the crowd says it is at that kind of work.
| THE JOB | BARE MINIMUMMac Mini M2 (24GB Unified Memory) | SWEET SPOTNvidia GeForce RTX 3090 24GB | FULL POWERMac Studio M2 Ultra (128GB Unified Memory) |
|---|---|---|---|
| Write an emaila paragraph or two, drafted from a one-line brief | ~42 S★★★★★ | ~14 S★★★★★ | ~17 S★★★★★ |
| Summarize a documenta long report boiled down to the points that matter | ~2.1 MIN★★★★★ | ~43 S★★★★★ | ~50 S★★★★★ |
| Build a websitea small landing page, markup and styles together | ~11 MIN★★★★★ | ~3.8 MIN★★★★★ | ~4.4 MIN★★★★★ |
| Build a backendan API with routes, storage and tests | ~35 MIN★★★★★ | ~12 MIN★★★★★ | ~14 MIN★★★★★ |
| Build a gamea playable browser game in one file | ~21 MIN★★★★★ | ~7.1 MIN★★★★★ | ~8.3 MIN★★★★★ |
STARS ARE THE COMMUNITY SCORE FOR THAT KIND OF WORK, DOCKED FOR HOW SQUASHED THE WEIGHTS GET AT EACH TIER. TIMES ARE COMPUTED FROM RESEARCHED SPEEDS. BALLPARK, NOT GOSPEL.
WHAT THE INTERNET SAYS
Users love Qwen3.8 27B for its fast speeds, strong coding help, and ability to handle long chats on consumer hardware. However, some people complain that it occasionally hallucinates, refuses prompts, or suffers from reasoning glitches.
9874 POSTS · 11281 COMMENTS · REDDIT · X · GITHUB · HACKER NEWS · LEMMY · DEV.TO · BLUESKY · BLOG · DEV FORUMS
PEOPLE LOVE
- Popular for coding help (71 mentions)
- People feel it is worth the money (34 mentions)
- Answers come back fast (32 mentions)
- Keeps up in long chats and big documents (9 mentions)
PEOPLE COMPLAIN ABOUT
- Refuses or lectures more than people like (124 mentions)
- Outages, errors, or limits get in the way (81 mentions)
- Sometimes makes things up (24 mentions)
TOPIC BY TOPIC
- Saying no too muchA SORE SPOT
1 thumbs up · 124 thumbs down
- Help with codeMOSTLY POSITIVE
71 thumbs up · 30 thumbs down
- Working when you need itA SORE SPOT
12 thumbs up · 81 thumbs down
- PriceMOSTLY POSITIVE
34 thumbs up · 20 thumbs down
- SpeedMOSTLY POSITIVE
32 thumbs up · 14 thumbs down
- Getting facts rightA SORE SPOT
15 thumbs up · 24 thumbs down
- Long conversationsMOSTLY POSITIVE
9 thumbs up · 5 thumbs down
- WritingMOSTLY POSITIVE
4 thumbs up · 2 thumbs down
- Doing what you askMOSTLY POSITIVE
2 thumbs up · 1 thumbs down
IN THEIR OWN WORDS
- FROM REDDIT >
“Local uncensored Opus 4.6 at home - Qwen3.8 27B heretic.”
- FROM X >
“Qwen 3.8 27b unrestricted whatever on my Spark has been grinding on a Jscript implant for 8 hours now 💪.”
- FROM REDDIT >
“Reasoning can be broken for some altered qwen3.8 27bs.”
- FROM REDDIT >
“Qwen3.8 27B is expensive than Qwen3.8 Flash and GLM5.3 Flash, is it a mistake?.”
- FROM X >
“I am so impressed by Qwen3.8 27B on Strix Halo...it is thinking a lot and at ~14t/s it is quite slow, but Jimmy has been.”
- FROM X >
“Tried Qwen3.8 27b with dflash2 on my strix halo, but the results are disappointing, not meaningfully faster than MTP.”
- FROM REDDIT >
“Qwen3.8 27B hallucinates user instructions during thinking.”
LOOKING FOR AN ALTERNATIVE?
Checked picks first: tags say why you would switch, stars say how fully each one stands in for Qwen3.8 27B.
Gemini 3.7 Flash
DeepSeek V4 Flash Vision Exp
Grok 4.6> SOURCES (12)
- REDDIT · 2026-07-19 · 4099 UPVOTESQwen3.8
Qwen3.8
- X · 2026-08-19 · 1489 UPVOTESWe’re releasing new Qwen3.8-27B GGUFs with 10% higher accuracy. Unsloth Dynamic V3 outperforms others by >10% on Div-
We’re releasing new Qwen3.8-27B GGUFs with 10% higher accuracy. Unsloth Dynamic V3 outperforms others by >10% on Div-
- REDDIT · 2026-08-03 · 1357 UPVOTESQwen 3.8 morning to you too Dario, 2$ input/ 6$ output per 1M.
Qwen 3.8 morning to you too Dario, 2$ input/ 6$ output per 1M.
- X · 2026-08-20 · 1308 UPVOTES💥 OBLITERATION ALERT 💥 ALIBABA: PWNED 🤗 QWEN-3.8-27B: OBLITERATED ⛓️💥 0.0% REFUSAL RATE across 842 harmful prompts
💥 OBLITERATION ALERT 💥 ALIBABA: PWNED 🤗 QWEN-3.8-27B: OBLITERATED ⛓️💥 0.0% REFUSAL RATE across 842 harmful prompts
- REDDIT · 2026-08-17 · 1127 UPVOTESArtificial Analysis' Qwen3.8-27B benchmarks put it neck and neck with DeepSeek V4 and GPT-5.6 Luna Max
Artificial Analysis' Qwen3.8-27B benchmarks put it neck and neck with DeepSeek V4 and GPT-5.6 Luna Max
- REDDIT · 2026-07-19 · 1059 UPVOTESJUST IN: Qwen 3.8 is coming. Open weight storm from China is continuing.
JUST IN: Qwen 3.8 is coming. Open weight storm from China is continuing.
- REDDIT · 2026-08-14 · 1013 UPVOTESQwen3.8-27B is out now!
Qwen3.8-27B is out now!
- X · 2026-08-16 · 995 UPVOTESMy honest takeaways after using Qwen3.8-27b on real world tasks for a few days: Grok-4.6 and Muse Spark-1.2 are definite
My honest takeaways after using Qwen3.8-27b on real world tasks for a few days: Grok-4.6 and Muse Spark-1.2 are definite
- REDDIT · 2026-08-17 · 989 UPVOTESAfter pushing 1M+ tokens through Qwen 3.8 27B, here is my optimal llama.cpp config for 16GB VRAM (73k Context, Agentic Coding)
After pushing 1M+ tokens through Qwen 3.8 27B, here is my optimal llama.cpp config for 16GB VRAM (73k Context, Agentic Coding)
- REDDIT · 2026-10-11 · 922 UPVOTESBuilding a 4x R9700 setup for a 10 person startup
Building a 4x R9700 setup for a 10 person startup
- X · 2026-08-15 · 893 UPVOTESQuick tip: download the open models you care about from Hugging Face as soon as you can. You never know what the next fe
Quick tip: download the open models you care about from Hugging Face as soon as you can. You never know what the next fe
- REDDIT · 2026-08-14 · 880 UPVOTESLocal uncensored Opus 4.6 at home - Qwen3.8 27B heretic
Local uncensored Opus 4.6 at home - Qwen3.8 27B heretic