This is the Pelican Test using Qwen 3.8 W4A16 Flash Next running locally.
Looks pretty good to me. What do you think?
Qwen 3.8 Flash Next ROCmi4 built this with OMP on AMD Strix Halo and I am so happy to make a fully working quant of this.
This works in Hermes + OMP and other coding harnesses. It built this retro cube game PLUS enabled it over my tailscale for me.
Ti è mai capitato che l'AI si rifiutasse di fare quello che le avevi chiesto?
Succede: ChatGPT, Gemini e Claude hanno regole interne che bloccano le richieste rischiose.
Qwen 3.8-FLASH is free right now
We can use Alibaba’s new 125B model for free through Qoder until September 30.
Just added 3 new fully private models to Wisp
Qwen 3.8 • Muse Glimmer 30b • Qwen 3.8 Uncensored All models run in a hardware-secured enclave so nobody can see, store or train on your prompts.
Be honest. How much of your "AI workflow" is just you copy-pasting?
Ask the chat for code → paste it into an editor → hit an error → paste the error back → repeat.
Qwen 3.8 Omni Flash can analyze the parts of your business that normally get ignored.
Calls, videos, meetings, images, and documents can all go into one workflow. The basic loop: → Feed in audio, video, text, and images.
Proudly present PAW-27B-X3, 3.000 people download this from huggingface.
Compressed model from Qwen 3.8 27B at 3.5bpw only 11.5GiB fit on single GPU RTX 3090 at full context 256k.
Et h a b e m u s ... C i n e m a g r a m !!
First tests !! CompfyUI + ChatGPT6 (make that json for ...) + MBP M3 max 128 GB Next : add sound, then voice, then lipssync, then improve quality,…
There's now an AI that watches a two-hour video the way you would.
It sees the screen. It hears every word. And it never loses the plot.
Alibaba just quietly shipped an AI that watches video like a person — start to finish.
No transcripts. No screenshots. No doing half the work yourself. It's called Qwen 3.8 Omni Flash.
Kamu: butuh jawaban lengkap buat uji sistem.
Model: "Maaf, saya tidak bisa membantu dengan itu." 🙃 Qwen 3.8 27B Uncensored: uncensored penuh, 131K context, chat/coding/agent, cache read Rp…
Qwen 3.8 omni realtime 日本語の声は
We ran Gemini 3.7 Flash, Gemini 3.8 Flash and self-hosted Qwen3.8-27B through the same grounded video QA pipeline on 300 Perception Test questions.
Gemini wins end to end, Qwen lands at 84% of its HOTA. The biggest lever was not the model at all. Write-up:
Try out cache-to-cache here
ONE DGX Spark Working with MiaAI_lab recipe of Qwen 3.8 flash next, but you can use any model Simple test of an essay about the research paper:
No single-stream headline today, so here's the part builders actually argue about
Two boxes running the same Gemma 4 12B IT QAT at 4-bit, and the spread between them is 3%. 29 tok/s on Strix Halo (Q4_K_XL).
Qwen 3.8 27B doing an agentic task with subagent at ~120 tok/s on M5 Max MacBook Pro in lmstudio Bionic using inco_ai Splash engine
At launch a little more than a month ago the model was running around ~20tk/s It’s 6x faster now, local AI is progressing at lightning speed!
Like any distilled model, Ternary Bonsai 2 27B inherits refusals and censorship from its teacher model: Qwen 3.8 27B.
Fortunately, there’s a very simple and elegant way to reduce or perhaps even completely eliminate them. 🧵
Your US stock position still carries risk even after your broker closes.
So I built Tesrune for the Bitget_AI S2 Hackathon. Tesrune is a dark-hours AI trading desk for US equities. You add the stocks you're holding.