The Archive

Meituan releases LongCat-Video-Avatar 1.5, open-source audio-driven human video generation framework with AT2V, ATI2V, and video continuation tasks.

LatitudeGames releases Equinox-31B, a Gemma-based 31B finetune blending dark adventure and slice-of-life storytelling data.

Cohere releases Command-A-Plus-05-2026 bfloat16 model weights on Hugging Face Hub.

Hugging Face releases Carbon, open DNA foundation models; Carbon-3B matches Evo2-7B SOTA while 275x faster with adapted LLM training for genomic data.

Introduction of Ettin Reranker Family models on Hugging Face.

Hugging Face reviving PapersWithCode with AI agents to auto-parse papers and generate leaderboards; covers Qwen, MTEB, ASR benchmarks.

Qwopus3.5-9B-Coder-GGUF: 9B dense model optimized for agentic coding and tool calling, runs at 8-bit on 16GB consumer hardware.

SupraLabs launches open-source small language models via Hugging Face, starting with Supra-Mini-v4-2M.

InternLM releases Intern-S2-Preview, a 35B multimodal model achieving comparable performance to trillion-parameter peers via task scaling in scientific domains.

inclusionAI releases Ring-2.6-1T, 1T-parameter open-weights reasoning model targeting agent workflows, engineering, and enterprise automation.

Hugging Face releases ml-intern, an agent framework for local LLM research automation supporting llama.cpp/ollama with Qwen and Claude models.

Resemble AI releases DramaBox, a voice synthesis model built on LTX 2.3, with open weights on Hugging Face.

Ovis2.6-80B-A3B multimodal model adopts MoE architecture for lower serving cost with improvements in long-context, high-resolution, and document understanding.

Hugging Face releases physics-intern, a multi-agent framework for theoretical physics research that doubles Gemini performance on CritPt benchmark.

Community milestone: 1M datasets published on Hugging Face, celebrated as progress for open-source AI.

MiniCPM 4.6 released on Hugging Face; open-weights efficient model variant with updated capabilities.

GGUF model uploads on Hugging Face nearly doubled in two months, signaling growth in quantized model ecosystem.

Community discussion of running MiMo-V2.5-GGUF model locally.

Hugging Face co-founder claims Qwen 3.6 27B matches Claude Opus on coding tasks offline, signaling convergence in local model capability.

Malicious infostealer masquerading as OpenAI privacy filter detected on Hugging Face; uses Python dropper to deploy PowerShell-based payload.

Zyphra releases ZAYA1-8B, an 8B parameter open-weights model claiming high intelligence density; available on Hugging Face.

Analysis of 100 most popular hardware configurations for local LLM inference on Hugging Face reveals deployment patterns and infrastructure preferences.

I built [hfviewer.com](http://hfviewer.com), a small tool for visually exploring Hugging Face model architectures. You can paste a Hugging Face URL and get an **interactive visualization** of the architecture, which can make it easier to understand how different models are structured and compare them at a glance. Here is the recent **Qwen3.6-27B** model as an example: [https://hfviewer.com/Qwen/Qwen3.6-27B](https://hfviewer.com/Qwen/Qwen3.6-27B) And here is a side-by-side view of the **Gemma 4** family: [https://hfviewer.com/family/gemma-4](https://hfviewer.com/family/gemma-4) Feel free t...

gemma-4-31B-it-DFlash open-weights model released on Hugging Face, pending llama.cpp integration.

8.7k synthetic fine-tuning dataset extracted from Claude Opus 4.6/4.7 outputs released on Hugging Face with reasoning traces.

inclusionAI releases Ling-2.6-1T, open-weight 1T-parameter model with MLA/Linear Attention hybrid for inference efficiency and coding tasks.

Mistral Medium 3.5 128B model released on Hugging Face.

Mistral releases Mistral Medium 3.5, a 128B dense model with 256k context window replacing Medium 3.1 and Magistral for instruction, reasoning, and coding tasks.

Ling-2.6-flash model released on Hugging Face; sparse details provided.