Open source, MIT, runs on CPU or any ggml GPU backend, and the smallest model is a 6 MB file that runs on a Raspberry Pi.
It's wired into @LocalAI_API too: a REST endpoint and live sound tagging over the realtime websocket, so you can stream audio and get tags as they happen.
Code, models and the benchmarks, MIT licensed: https://t.co/3GEpfVSiSq
How well do LLMs migrate pandas to Polars by themselves?
We tested how well Claude translates a pandas corpus to Polars. Results were promising but not perfect. To improve this, we built a Polars skill that helps the agent.
Read the full post here: https://t.co/UPTkYVNkCi
ATTENTION: BEFORE YOU SHARE THE CLAIM THAT "ROMANIA IS ADVANCING A BILL TO UNIFY WITH MOLDOVA," READ THIS. Right now, this story is being used by the Kremlin to push the narrative that Romania wants to "annex" Moldova. Don't amplify the narrative without the full context. ⬇️🧵
Hampton Court last Friday night was so special, I’m still dreaming about it and unusually for me, I think I can still remember all my fun facts about the venue as this place had facts a-plenty! 🩷✨ What a beautiful venue - can every gig be in a pleasure palace please 👑? Xxx
🚀 The 7-Day Voice AI Builder Challenge is Officially LIVE!
Stop babysitting your terminal.
🗣️ The Challenge: Teach your AI coding agent to call you for backup—but only when human intervention is actually required.
Real-time feedback? Yes. Live leaderboard? Absolutely. Epic prizes for the winners? You bet.
The clock is officially ticking. Competitors are already shipping.
Are you in?
👉 https://t.co/NFrXKoiprV
This model diverges from the previous LocalVQE models being based on GTCRN instead of DeepVQE. It differs from GTCRN in that it has AEC. The quality is not the same as the larger LocalVQE models, but it is orders of magnitude smaller. You can compare the different versions at: https://t.co/erFTqk7BpT
Welcome to the diffusion era.
We bet on parallel generation years ago, when it was a contrarian idea. It's great to see the industry arrive.
Mercury 2 continues to lead the Pareto frontier for quality, speed, and cost among publicly available diffusion LLMs.
More than 40 years ago, I arrived in Chicago in search of an idea. I was a young man looking for purpose, who believed deeply in America, was inspired by the Civil Rights Movement, and wanted to be a part of something larger. The America I believed in was one where everyone has opportunity, everyone is seen, everyone belongs—because that was an America that had a place for me, too.
WHAT THE HELL is happening in AI?
A 3B parameter model just put up coding benchmark scores in the same league as Claude Opus 4.5.
3 BILLION.
The weights are on Hugging Face, anyone can test it.
I genuinely don't know if this is a breakthrough or if the benchmarks are broken.
Want to teach Gemma to master chess?
Check out this awesome community project showing how to fine-tune Gemma 4 12B on your own data, 100% locally!
Running text, images, and audio on just 8GB VRAM makes custom models more accessible than ever.
From a generative 3D Earth to agentic final exams — the most upvoted AI papers this week
Top papers on @huggingface Daily Papers (June 8–14):
- ABot-Earth 0.5: Generative 3D Earth Model
- Agents' Last Exam
- Kwai Keye-VL-2.0 Technical Report by Kwai
- MiniMax Sparse Attention by MiniMax
- EvoArena: Tracking Memory Evolution for Robust LLM Agents in Dynamic Environments
- Imaginative Perception Tokens Enhance Spatial Reasoning in Multimodal Language Models
- SWE-Explore: Benchmarking How Coding Agents Explore Repositories
- Toward Generalist Autonomous Research via Hypothesis-Tree Refinement
- WeaveBench: A Long-Horizon, Real-World Benchmark for Computer-Use Agents with Hybrid Interfaces
- Your UnEmbedding Matrix is Secretly a Feature Lens for Text Embeddings
Advice for AI coders 💡
Frontier coding models are very useful, but they can also (and surely will) get super expensive.
Agentsview is tool to help you track costs and usage, so you can pick the right tradeoff between quality and cost.
Enjoy ⬇️
https://t.co/LEWkgwvfYT
@mx_alxr "... validated that the level of capability displayed there is widely available from other models (including OpenAI’s GPT-5.5), and is used every day by the defenders who keep systems safe." Это особенно прекрасно, т.е. мы вам продавали, но то же можно и в 5.5 сделать.