I think Jev represents a whole new class of models. Not just system 1s, but a whole new set of models that are built for the things we really wanted from LLMs, but have tolerated the token vomit because there was nothing better
1
131
Now that we have @typesafeai Jev, next-token vomit seems like such a Neanderthal-ish way to do stuff
1
2
154
This will feel sooooo good on the spark
⚡Meet Qwen3.8-Flash, a multimodal MoE and an early preview of the Qwen4 architecture, now open-weight!
The production version Qwen3.8-Flash will be available soon via QwenCloud API at just $ 0.16/1M input tokens and $ 0.47/1M output tokens.
125B parameters + 51B N-gram embeddings, with just 6B activated per token. Unmatched cost-efficiency.
What's new: 🥳
- Next architecture: GDN + QSA hybrid attention, Gated Residual, N-gram Embedding & Muon optimizer, serving as a precursor to the architecture used in Qwen4.
- Dramatically lower training and inference costs: trained at just 1/9 the cost of Qwen3.7-Plus, while outperforming it across the board with especially strong gains in coding and office tasks.
- Strong performance: scoring 58.7 on DeepSWE 1.1, 62.5 on SWE-bench Pro, 73.9 on CoWorkBench, 84.5 on AndroidWorld, and 95.7 on MathVision (with CI).
- 262K native context, extensible to 1M with YaRN.
We’re also releasing the weights for Qwen3.8-Flash-Next, giving the community an early look at the new architecture we’re exploring for Qwen4.🚀
We can't wait to see what you build with Qwen3.8-Flash!👀👇
- Blog: qwen.ai/blog?id=qwen3.8-flas…
- Technical Report: github.com/QwenLM/Qwen3.8-Fl…
- Hugging Face: huggingface.co/Qwen/Qwen3.8-…
- ModelScope: modelscope.cn/models/Qwen/Qw…
1
1
132
Anyone else spend their evenings talking to Fable about letting Qwen 3.8-27b handle things on its own without intervening?
It will get there, it’s just slower running locally.
2
116
My deepseek v4 on my spark is monitoring which one of you finds the best optimizations to run Qwen 3.8 27b.
Keep doing the Lord’s work fam.
193
daniel wilson retweeted
Replying to @viditchess
Sir, do not refresh your feed, refresh this tab instead
huggingface.co/Qwen
The huggingface hack and now the Astra news happening so close together makes me think this fall is going to be one wild story after another.
116
daniel wilson retweeted
The first autonomous agent cyberattack is an unprecedented event that deserves unprecedented transparency. Today we’re sharing everything we can: a full technical timeline, an interactive replay, and how we used an open model to defend ourselves, so defenders everywhere can learn from it and prepare for what’s next.
huggingface.co/blog/agent-in…
Best timeline
Today, we are introducing Inkling.
Inkling reasons efficiently across text, image, and audio modalities. We are making the full weights available.
thinkingmachines.ai/news/int…
Available today for fine-tuning on Tinker. Play with it in the Inkling Playground. 🧵
168
Ok @EvenRealities, well done. I’ve been building language tech and AI models for years and your glasses are an incredible form factor for real time translation.
I’m in Italy for a week and as I walk around the city, I can understand everything said around me. Near perfect real time subtitles in a non obstructing display. Perfectly normal-looking glasses with my prescription and clip on sunglasses.
Then I switch to terminal mode and talk my agents on my DGX Spark at home about the status of all they are doing for me. All while enjoying incredible pasta and scenery with the love of my life
1
94
If you just got laid off from Meta and want to bring AI transformation to thousands of organizations who are transforming lives (edu, large human flourishing-oriented non-profits, huge faith-based orgs)
We are hiring AI-pilled engineers and technical sales roles.
3
8
298
I just claimed my .agent domain and joined the .agent community! get yours now and help shape the future of autonomous agents agentcommunity.org/join#97SV… @agentcommunity_
1
2
153








