Released last week, and already more than 4M downloads on HuggingFace alone 😊 This makes Gemma 4 12B the most popular encoderfree VLM by a large margin. In addition to being the first-ever general purpose LLM with encoderfree audio input!

I'm seeing a lot of angry people lately... remember, you can always run your coding agent locally ;) llama.cpp OpenCode = fast, reliable and private inference. This is @UnslothAI North-Mini-Code-1.0-GGUF running at ~50 tokens/s on my Macbook

Alibaba Qwen3.7 slowly fading into irrelevance at the frontier due to proprietary stance. In it's place we have Minimax M3 and... *checks notes* Rio 3.5 397b, made by the municipal IT company of Rio de Janeiro's city government. huggingface.co/prefeitura-ri…

Today we're releasing ZONOS2, our next-generation real-time TTS model with high-fidelity voice cloning. ZONOS2 is the most expressive open-source TTS model, released under Apache 2.0 and available on Zyphra Cloud on @AMD. 🧵

SITUATION DETECTED: The city of Rio de Janerio has post-trained a model. Based on Qwen 7/2, Rio 3.5 Open 397B adds SwiReasoning on top of the base Qwen model — a framework that dynamically switches between standard chain-of-thought and latent-space reasoning, guided by entropy-based confidence signals, so the model only "thinks out loud" when it needs to and otherwise reasons silently in hidden space for better token efficiency.

There is no inevitability in AI. We all have agency in what comes next: Path 1: closed-source APIs, concentration of power, and a future decided by a handful of people in Silicon Valley and DC Path 2: open-source AI, where everyone gets to participate, own, and build together, including orgs like the city of Rio. Pick your path anon!

MiniMax v3 is out and it ships with an MSA kernel that gets crazy speedups the longer your sequence length is. It's shipped on the Kernel Hub as well and is integrated in transformers: huggingface.co/kernels/MiniM…

interesting MiniMax shipped a kernel on Hugging Face 👀 huggingface.co/kernels/MiniM…

🤗 MiniMax M3 from @MiniMax_AI is now live on @huggingface — supported by Novita. Open weights. ~428B total parameters. ~23B activated parameters. Built for the Agent Era.

Have a great week end (don't forget to touch grass 🍃)

new transformers tutorials just dropped for vision 🔥 🛰️ segmentation on satellite imagery: fine-tune RF-DETR-Seg segment buildings 📱 object detection on mobile UI: fine-tune RF-DETR on screenshots runs on toaster, converges fast, give to your agent for your use cases🫡

MiniMax M3, Open-Weight, Now On Hugging Face , with only ~428B parameters and ~23B activated parameters Weights: huggingface.co/MiniMaxAI/Min… MiniMax Sparse Attention: huggingface.co/papers/2606.1…

🔥🔥🔥 huggingface.co/moonshotai/Ki…

Published my first kernel to go the last mile to optimize LTX-2.3 from @Lightricks! torch.compile cuDNN attn already gave a 1.42x boost. W/ the custom kernel added, I got 1.52x on a GB10 🔥 This was my systematic exploration of a simple agentic kernel dev workflow. More 👇

Real-time social robotics, from the cloud to your local device. Watch Ian from our DevX team use Gemini Live for a seamless voice chat with Reachy Mini. Then, stick around until the end to see the robot running locally on Gemma 4!

Agents are only as good as the environments behind them. At Mercor, we've built deep expertise in the realistic, economically-grounded environments that help agents bridge the gap from the lab to real-world usefulness. We want to put that expertise to work for the broader ecosystem—so we're glad to be joining the OpenEnv committee, alongside Meta @PyTorch, @nvidia, @PrimeIntellect, @huggingface, and others, to help guide the open foundation for agentic environments.

HF has become the best storage platform for PRIVATE and PUBLIC models and datasets, both intermediary and final ones! Great example from @heyjasperai who used HF buckets to store their Monet dataset and train models directly on it! More details: huggingface.co/storage/testi…

Explore your @huggingface repos in a whole new way 🔥 Visualize storage, discover outliers, and navigate your repos directly from the terminal. `hf repos ls --explore`

Build Small with Modal x.com/i/broadcasts/1yKAPPvOX…

🔓 And the best part — we're open-sourcing it. 1,000 tps on a 1T model wasn't a single breakthrough — it's deep model × system co-design between the MiMo and TileRT teams, all on general-purpose GPUs (no Cerebras-style wafer-scale, no Groq-style SRAM ASICs). On the model side: FP4 quantization (smaller footprint, less memory traffic) DFlash, our block-masked parallel speculative decoding that accepts far more tokens per verification. On the system side, TileRT tailors its compiler & kernels to exactly these techniques. The result: a 1T model breaking 1,000 tps on a single, standard 8-GPU node. 🤗 Open weights (FP4 DFlash checkpoint): huggingface.co/XiaomiMiMo/Mi…