Qwen 3.8
Alibaba's open-weight language model family with 27B and Flash Next variants.
HN commenters run it locally on consumer GPUs and Macs as a daily driver12345, praising its coding ability for its size.6789 The most common complaint is long reasoning that makes hard tasks slow10111213, with reliability dropping at low quants.1415
Why people recommend it
Runs locally · 3
Commenters run it on single consumer GPUs, including 16GB and 24GB cards, and on Macs.
“I have been using qwen3.8_27B_UD_Q3_K_XL. This does 46 tok/s on a 5080 with 16gb vram and it can pretty much do any systems work”
“I use it at 60+ tok/second on a single 24GB card.”
Show 1 more
“I run Qwen 3.8 27B just fine on my Mac mini M4 24GB.”
Coding ability · 3
Users find it strong for software development, with some saying it beats DeepSeek V4 Flash at coding.
“Qwen is faster and seems to be much better at coding tasks.”
“We recently moved from 27B to Flash Next. The quality is superior for coding.”
Show 1 more
“qwen3.8 is more than capable for software dev.”
Daily driver · 3
Several commenters say it is the main or only model they use day to day.
“I use Qwen3.8-27B as a daily driver”
“since 3.8 27b came out it is all I use”
Show 1 more
“Qwen3.8-27B is a huge step up from Qwen3.6-27B.”
Fast throughput · 3
Users report high token rates on local hardware, especially with Flash Next and multiple streams.
“My server can run Qwen Flash Next at 500 t/s aggregate (about 200 t/s single user). It's pretty great.”
“qwen3.8 27b is different.”
Show 1 more
“I am just so impressed with Qwen Flash Next right now. I have been using it locally and it's just such a great little model to work with.”
Near-frontier quality · 3
Some compare it to Sonnet or Opus for coding at far lower cost, though slower.
“I've been running it locally since for real programming and I consider it pretty much the same as opus 4.6 in coding ability”
Show 1 more
“achieving flagship quality without the price tag (albeit slower)”
Long context · 3
Commenters report working sessions up to 200k+ context without noticeable degradation.
“qwen3.8 27b is different.”
Show 1 more
“one of the cheapest ways to get qwen3.8 27B Q6 with 100k+ context working”
Caveats and dissent
Overthinking · 3
It can reason for very long stretches, sometimes never finishing or taking much time and many tokens.
“The main issue i found is that Qwen3.8-27B thinks a lot. On a hard problem the planner runs 32K tokens of reasoning and never writes the plan.”
“Quanted out Qwen 3.8 models are not frontier level. Have you used them for real work? They suck.”
Show 1 more
“It's the first small local model I've felt like I can do real work with.”
Quantization losses · 2
Low-bit quants are described as less reliable and prone to looping or inconsistent results.
“Quanted out Qwen 3.8 models are not frontier level. Have you used them for real work? They suck.”
“3.8 is absolutely quite a bit better than 3.5.”
Slow on Macs · 2
Some Apple Silicon users found the 27B model too slow to use comfortably.
“It's just too slow of a model.”
“Qwen3.8 wasn't fun to use on my 64GB M4 Max either”
Nuanced tasks · 2
Some find it falls short of frontier models on complex or subtle work.
“I have found qwen to be absolutely useless at anything but very specific, surgical code changes.”
“Qwen3.8 27B does not seem to be good enough to implement the simulator with all its intricacies and tiny subtleties correctly”
World knowledge · 2
Commenters note weaker factual knowledge and advise giving it tools for context.
“I've been running it locally since for real programming and I consider it pretty much the same as opus 4.6 in coding ability”
“Qwen 3.8 27b in my experience is more than capable of churning out features overnight with the right tools”
Often compared with
1 moved from Claude Opus to Qwen 3.8
1 moved from DeepSeek V4 Flash to Qwen 3.8
2 moved from GLM to Qwen 3.8Pi15 comments mention both
Who recommends it in their own words
Threads where it came up
- Benchmarking Qwen3.8 27B quantizations: 4-bit holds up, 1-bit collapses 7 mentions
- Ask HN: What default model do you use and why? 6 mentions
- Qwen 3.8 27B is excellent, but it defaults to overthinking things 6 mentions
- Qwen 3.8 27B is out: open weights, best local dense model yet 5 mentions
- Qwen3.8-Max: A New Bar for Coding and Cowork 5 mentions
- Run Qwen 3.8 Flash Next (125B) on consumer hardware (RTX 4090) at 100T/s 4 mentions
Also written as Qwen 3.8, Qwen3.8 27B, Qwen3.8-27B, Qwen 3.8 Flash Next. Data: qwen-3-8.json. Quotes are verbatim; the verdict and theme labels are written by a model from these comments, and every clause links to its sources.
All comments 44
“It farms smaller sub tasks to a fleet of 7 Qwen3.8-Flash-Next agents.”
“I use Qwen 3.8 27b, yeah.”
“one of the cheapest ways to get qwen3.8 27B Q6 with 100k+ context working”
“My server can run Qwen Flash Next at 500 t/s aggregate (about 200 t/s single user). It's pretty great.”
“The main issue i found is that Qwen3.8-27B thinks a lot. On a hard problem the planner runs 32K tokens of reasoning and never writes the plan.”
“they have gotten quite good but still make noticeably more hallucinations or bad assumptions”
“My use of Qwen 3.8 27B (which is the only model i use almost since it came out)”
“Qwen 3.8 27b on my 5090, barely acceptable, thinks too long, sometimes freaks out.”
“I use it at 60+ tok/second on a single 24GB card.”
“I am just so impressed with Qwen Flash Next right now. I have been using it locally and it's just such a great little model to work with.”
“Both were great, but surprisingly, Q3.8FN had the better result.”
“Qwen is faster and seems to be much better at coding tasks.”
“I switched to Qwen for image input”
“Yes 27B - not so much of a surprise but I was hoping for a usable coding model that's local and fast”
“locally hosted Qwen-3.8-27b or Qwen-Flash-Next”
“locally hosted Qwen-3.8-27b or Qwen-Flash-Next”
“Quanted out Qwen 3.8 models are not frontier level. Have you used them for real work? They suck.”
“if you're VRAM constrained, take a look at GLM 5.3 Flash or Qwen 3.8 Flash Next”
“I have been using qwen3.8_27B_UD_Q3_K_XL. This does 46 tok/s on a 5080 with 16gb vram and it can pretty much do any systems work”
“Qwen 3.8 27b / Gemma 12b”
“feeding the results through local Qwen 3.8 running under MTPLX”
“I use Qwen3.8-27B as a daily driver”
“My tests with Qwen 3.8 have convinced me it's completely fine general purpose”
“I use Qwen 3.8 26b at home and find it very useful for some tasks”
“I use MiMov2.6Pro, DeepSeekv4.1Flash, GLM5.3, Hy4, Qwen3.8 and KimiK3 at home.”
“this model is the first local model that does a good job. It is kind of amazing.”
“3.8 is absolutely quite a bit better than 3.5.”
“We recently moved from 27B to Flash Next. The quality is superior for coding.”
“I use it with the browser that has a great dashboard and with hermes agent and that it really makes this amazing.”
“I'm working on local models; qwen3.8-Flash-Next seems to be passed to rubicon.”
“the results are so very much excellent enough for about 60% of my work”
“That's good enough for me for development.”
“either a blazing fast 8-bit Qwen 3.8 27b or a 4 or 5 bit Qwen 3.8 Flash Next”
“when using Qwen 3.8 flash next”
“qwen3.8 is more than capable for software dev.”
“qwen3.8, kimi3, kimi2.7, GLM-5.3 are all good families I use in my coding team”
“I've been running it locally since for real programming and I consider it pretty much the same as opus 4.6 in coding ability”
“I use it with a local Qwen3.8 27B on llama-server”
“achieving flagship quality without the price tag (albeit slower)”
“Local model is Qwen3.6 35B-A3B and Qwen3.8 27B UD.”
“Qwen 3.8 has been the first local model I've tried that has consistently given usable and impressive results.”
“Qwen3.8 27B does not seem to be good enough to implement the simulator with all its intricacies and tiny subtleties correctly”
“since 3.8 27b came out it is all I use”
“It's the first small local model I've felt like I can do real work with.”