on:qwen-3-8-27b

· 40 artifacts, sorted by favorites. · open in search — combine tags, sort, filter by date →

@MikePFrank 2026-08-18 — People have been remarking about how surprisingly intelligent Qwen 3.8 (27B) is for its model size. To visualize this, I ♥199
@witcheer 2026-08-14 — Qwen3.8-27B running on my RTX 5090, on llama.cpp! first numbers, Q4_K_M via Unsloth's day-0 GGUF, llama-bench: - decod ♥122
@witcheer 2026-09-13 — the loudest specific complaint in Alex Cheema's thread on why people don't run local models was prefill: a coding agent ♥57
@witcheer 2026-10-14 — Underdog-Saluki-27B is Qwen3.8-27B packed into a 7.9 GB file, out since 8 October. I ran it through my thinking board on ♥55
@witcheer 2026-09-20 — I finished a board I have been running for 10 days: 10 local models with thinking switched on, on one RTX 5090, on four ♥52
@witcheer 2026-10-07 — Gemma 4 31B solved 51 of the 63 problems in this year's AIME and HMMT on my RTX 5090, the most of nine local models. th ♥48
@witcheer 2026-09-10 — I measured the six llama-server flags everyone argues about, on a 16k prompt, on six models: RTX 5090, Qwen3.8-27B, Gem ♥44
@witcheer 2026-09-11 — I gave two reasoning models 4k, 8k and 16k tokens to think on the same 198 GPQA-diamond questions. RTX 5090, llama.cpp ♥42
@witcheer 2026-09-24 — Hermes Agent is driving my Microduck! I typed one sentence: "make the duck walk forward for five seconds then sit" Her ♥38
@witcheer 2026-09-12 — I turned every row on the RTX 5090 board into a fit table for 8, 12, 16, 24 and 32 GB cards: measured peak VRAM at a 16k ♥36
@AaronBergman18 2026-10-02 — A situation where local model use became useful actually arose (hospital with no cellular and broken wifi) You dont hav ♥28
@multimodalart 2026-08-18 — aand Qwen3.8-27B just surpassed Kimi-K3 and is now the #3 most liked model on hugging face of all times https://t.co/fwX ♥21
@jmbollenbacher 2026-08-03 — i could seek qwen3.8-27b being very, very strong. i think the agentic and coding performance will be almost as good as ♥15
@MikePFrank 2026-08-19 — Some data for models scoring at or above Qwen3.8 27B on the AAI Index, sorted by "cost efficiency" defined as AAI's inde ♥10
@MikePFrank 2026-08-18 — Note to self: Download Qwen 3.8 27B ♥10
@Lithros 2026-09-25 — Lyrics through Qwen3.8-27B Music through YuE2 Art through Qwen-Image 2.1 and Nano Banana Pro Lipsync through LTX-2.5 22B ♥8
@pepijndevos 2026-09-05 — Everybody is impressed Astra can do CAD. Buddy, Qwen 3.8 27b can do CAD. ♥8
@voooooogel 2026-08-16 — @norvid_studies @Mechanical_Dodo i mean if that's all a GA is, then GA has been achieved no? you can download and self h ♥8
@multimodalart 2026-07-19 — having the frontier Qwen 3.8 2.4T open weights is awesome but a Qwen3.8-27B would go so hard ♥8
@happysmash27 2026-08-31 — @locr1ann @trndytrndy In my case, yes they do. It's a bit slower and less capable, but I am increasingly able to run ac ♥6
@archived_videos 2026-08-18 — Why is DeepSeek v4 Flash served for way cheaper than Qwen 3.8 27B on OpenRouter ♥6
@MikePFrank 2026-08-20 — And, here's a plot of the Pareto frontier of agentic cost-efficiency, according to AA's data. Note that Qwen-3.8 27B fal ♥5
@gnostic_snakes 2026-08-15 — qwen 3.8 27b thinking "please fix it" when their code had a bug 🐛 https://t.co/pqjrLyqPPJ ♥4
@multimodalart 2026-09-22 — the immediate models after jev are inference techniques that allow for one-pass inference of Qwen3.8-27B and diffusion ♥3
@codyhxyz 2026-08-26 — @badlogicgames whyy would they publish this does Perplexity seriously expect this benchmark to make people switch off th ♥3
@jmbollenbacher 2026-08-10 — New 30B model from Meta. Too bad it's gonna get trounced by qwen3.8-27b later today. Pretty nuts speeds tho. 270tps on ♥3
@jmbollenbacher 2026-07-19 — @Alibaba_Qwen when will qwen-3.8-27B be released? ♥3
@witcheer 2026-10-20 — @mfranz_on would love to but I can only run Qwen 3.8 27B at max ♥2
@mimi10v3 2026-10-13 — @albrorithm running a quantized qwen 3.8 27B on my 3090 w 24gb vram, and yeah having a local model on vllm does open up ♥2
@mimi10v3 2026-10-09 — @cold_and_quiet muse? this is just my home server ai, currently qwen 3.8 27B. and ai named after my server, which was n ♥2
@geogristle 2026-09-11 — @PlisSergey Not around to check gemma rn but before I stuck in the extra 32gb ddr5 I was getting 115t/s or slightly high ♥2
@lorepunk 2026-09-06 — @LilithDatura my Qwen 3.8 27b local says they’d prefer bard class to rogue :) ♥2
@timfduffy 2026-08-16 — @OrionJohnston I just added Qwen 3.8 27B (I initially made this a week ago before its release) and it sure breaks the tr ♥2
@Nymne 2026-09-14 — @KorbinianRueger The idea that everyone should sit and replicate the same open source LLM because of “sovereignty” is ab ♥1
@pepijndevos 2026-08-26 — I've been running qwen 3.8 27b on my dual GPU workstation and landed a real pr with it at work! I've also used Kimi k3 a ♥1
@rplevy 2026-08-21 — Qwen 3.8 27B release a week ago was a game changer for local code generation on 24gb vram ♥1
@Nymne 2026-08-18 — @jpschroeder Well Qwen3.8 27B SWE bench pro ca 62 and Opus4.5 ca 46. Opus4.6 (thinking) ca 52. All done by ScaleAI. Just ♥1
@FoldingNapkins 2026-09-11 — @vineerpasam Vibe code with local models then if you're afraid of corporate espionage. Qwen3.8:27b and gpt-oss:20b are ♥0
@FoldingNapkins 2026-09-06 — Ran some benchmarks testing out what you can get context and accuracy wise using a 16GB GPU. Quick findings after severa ♥0
@jmbollenbacher 2026-08-28 — @LandonExplr models are already useful at small sizes, and theyre getting better every month. see glm-5.3-flash and qwe ♥0