A focused feed of models, agents, research and open-source releases for people building with AI.
Microsoft's BiomedBERT (formerly PubMedBERT) is trending with roughly 870K downloads on its current model page. Pretrained from scratch on PubMed abstracts and PubMedCentral full text, not adapted from general BERT.
LearnHarness is an open-source desktop agent that plans lessons from your current understanding to your goal: GitHub repo Four-stage loop: measure prerequisites, plan a dependency map, teach step by step, then practice Persistent knowledge graph stores propositions and reasoning chains, reused
Deep Principle released Hydrogen-1.0 , a 753B open-weight scientific model under MIT license. Post-trained from GLM-5.3 on chemistry, materials science, life-science data, and internal lab records. Scores 38.3% on FrontierScience-Research vs GPT-6-Astra's 33.3%, and 93.0% on GPQA-Diamond.
Rikiya Takehi released an edited Qwen3.5-0.8B on Hugging Face, modified by GPT-6 Astra without any training. Astra changed 0.016% of weights (131,072 values) across two MLP down-projection tensors in layers 15 and 19.
Pruna AI open-sourced P-H3-Image , converting MiniMax-H3 video model into an image generator. Generates 8-megapixel images in 8 steps, roughly 5x faster than the base model. Smallest w4a8 variant is a 20 GB download peaking at 21 GB VRAM, versus 143 GB for MiniMax-H3.
Agent Autoloop is a self-hosted middleware that drives CLI agents like Claude Code and Codex in the background. Install with a single curl command on Mac or Linux; no cloud server or public port required.
New paradigm, Anthropogenic Regional Adaptation , measures human-centric regional alignment in vision-language models. GG-EZ method combines regional data filtering, supervised fine-tuning, and weight merging to localize models cheaply.
Mesh Avatar Studio turns one illustration into an animated 2D VTuber rig, hitting ~1.8k GitHub stars under MIT license . A coding agent (Claude Code or Codex) places the mesh, cuts layers and reviews poses from docs/agent-guide.md .
OrcaRouter released Flint-abliterated-GGUF , an uncensored 40B MoE with 262K context. Two builds: 12 GB for 16 GB GPUs, 24 GB for 32 GB GPUs, Apache 2.0. Harmful-prompt refusal drops from 86% to 10% with MMLU barely moving. Mix-24G beats Unsloth UD-Q4_K_XL on KLD while being 1.1 GB smaller.
New framework reformulates RAG grounding as case + claim + evidence support verification. Builds training data with counterfactual wrong-state and topic-related negatives automatically. Verifier hits 97.4 AUROC on MIMIC-CXR , collapses to 55 under swapped evidence.
Community retrofit adds Qwen Sparse Attention to Qwen3.8-27B, previewing a core Qwen4 technique on an existing checkpoint. At 261k tokens, QSA delivers 2.9x faster prefill and 1.8x faster decode than dense.
Community quant repack of Qwen-Image-2.1-Turbo hits 2.5M downloads and 4.5k likes on Hugging Face. Eleven formats: GGUF (Q4 to Q8), FP8, INT8 ConvRot, NVFP4 safetensors, and MLX 4/6/8-bit for Apple Silicon. Base model generates in 8 denoising steps at CFG 1, versus 40 steps for Qwen-Image-2.1.
Google released MedGemma , open medical vision-language models built on Gemma 3 4B and 27B. Includes MedSigLIP , a 400M medical vision encoder usable standalone for classification and retrieval. Scores 86.2% on MedQA, best open model at release aside from DeepSeek-R1.
Artificial Analysis launched AA-PersonalityBench , grading 42 models on 30 personality traits and 21 writing characteristics. Top proprietary models beat top open weights on Truthful by 32 points (79 vs 47), the widest single-trait gap.
Artificial Analysis updated its six domain Capability Indexes to v2.0, aligning them with Intelligence Index v5 . Every index now weights Non-Hallucination and Agentic Tool Use more heavily, reflecting real deployment risks.
ffmpeg-skill gives coding agents 42 local FFmpeg tools with no cloud or API keys. Installs with one npx ffmpeg-skill command for Claude Code, Cursor or Codex. Enforces a probe, edit, verify workflow with contact-sheet PNG inspection by the agent.
New arXiv paper proposes a 3D cell tracking pipeline designed for sparse annotations: arxiv.org/abs/2611.06624 . Combines 2.5D U-Net and 3D SegResNet detectors with loss masking around likely unannotated cells.
Krea 3 Ultra debuts at #4 on Arena, Artificial Analysis, and Design Arena editing leaderboards . Finishes edits in ~8s, 3-5x faster than Nano Banana Pro and 11x faster than GPT Image 2.5. $80 per 1k edits vs $150 for Nano Banana Pro and $436 for GPT Image 2.5.
Partially decensored GGUF of Swift 1.5 Qwen3.8-27B built with Heretic , with vision and MTP intact. Scores 97.6% GSM8K, 86.67% MMLU-Pro, 92.98% IFEval prompt-strict on the quantized file. Follows through on roughly 1 in 5 prompts the original Swift 1.5 refused; not fully uncensored.
Artificial Analysis is exposing the CritPt v2 physics benchmark grading server through its API. CritPt covers 71 composite research challenges built by 50+ active physics researchers across 30+ institutions.
Anthropic released FlashProve , open-source cryptographic proofs that a response came from a specific LLM. Generates proofs for 400B Llama 4 Maverick at 110 tokens/second on a single H100 GPU. Proofs are ~19 MB and verifiable in seconds on an ordinary 24-core CPU without weights.
Shawn Pana open-sourced Phone Harness , a Python toolkit letting agents control iPhones and Androids. Works via iPhone Mirroring plus Apple Vision OCR on macOS, and adb plus the accessibility tree on Android.
Prime Intellect audited AutomationBench, OfficeQA Pro v2, and HLE-Diamond, finding broken graders and reward hacks that distort reported model scores. AutomationBench fixes replaced prose request bodies with JSON Schemas; Opus 5.5 hit 92.15% partial, 70.83% strict.
Topaz Labs released Wonder 4 , a one-click upscale, denoise, and sharpen model. First Topaz model built on a mixture-of-experts architecture that routes enhancements per image region. Combines upscaling, denoising, deblurring, compression artifact removal, and face recovery in a single pass.
No stories match these filters.
