
Cognition's Devin Restores Claude Fable 5 After a Messy Government Ban
Claude Fable 5 is back in Devin after a government-ordered suspension — now powering the Ultra agent across Cloud, Desktop, and CLI
20 stories

Claude Fable 5 is back in Devin after a government-ordered suspension — now powering the Ultra agent across Cloud, Desktop, and CLI

NVIDIA splits a 30B model into two specialized towers to generate text 2.42x faster while keeping 98.7% of the original model's quality

Google's new open-source VS Code extension lets you run Workbench Jupyter notebooks on cloud GPUs without leaving your local IDE

LlamaIndex ships legal-kb, an open-source reference app showing how to give AI agents filesystem-style tools to autonomously navigate large document knowledge bases

Cognition's Devin Security Swarm uses a new Agentic MapReduce architecture to find real vulnerabilities across entire codebases, outperforming Claude and Codex at 30% lower cost.

GitHub Issues now warns you about duplicate issues before you submit, and AI agents can fully triage issues via the MCP server

Zed v1.9 ships Telescope-style resizable pickers with live preview, in-thread agent search, and a new text finder modal — resolving an 860-upvote community request

Z.ai launches ZCode, a dedicated desktop IDE for GLM-5.2, with 1.5x quota for subscribers and BYOK support across all platforms

NVIDIA quantizes Mistral's 128B unified model to FP4, slashing memory by 1.8x vs FP8 with near-zero accuracy loss on Blackwell GPUs

Factory's Droid Shield 2.0 adds two fine-tuned LoRA models to catch secrets that rule-based scanners miss or wrongly flag, beating GPT-5.5 and Opus 4.8 on both tasks.

xAI's no-code Voice Agent Builder lets anyone deploy a production phone agent in two minutes, powered by the #1-ranked Grok Voice model, at $0.05/min

ebook2audiobook turns any DRM-free ebook into a chaptered audiobook with voice cloning, 1158 languages, and zero per-book cost

Databricks' LTAP architecture stores transactional and analytical data in a single open-format copy, eliminating CDC pipelines, ETL jobs, and data drift forever.
Fish Audio opens its best TTS model, S2.1 Pro, to all developers for free — 83 languages, voice cloning, no hard usage cap

Tencent's 1B-parameter HunyuanOCR beats models 200x its size on document parsing, spotting, and translation — and it's fully open-source

Tencent's MixGRPO cuts RLHF training time for flow-based image generators by up to 71% using a sliding-window ODE-SDE hybrid, accepted to ECCV 2026.

DimOS is an open-source Python framework that lets LLM agents drive real robots without ROS, wiring perception, SLAM, and control into one composable stack.

Together AI's ParallelKernelBench reveals frontier models solve under a third of 87 real multi-GPU kernel problems, exposing a critical blind spot in AI coding benchmarks.

Kilo Code's new Auto Efficient router automatically matches each coding task to the cheapest model that can handle it, cutting AI bills by up to 72% without sacrificing output quality.

Chinese researchers release an 8B open-source agent that runs the full data science pipeline autonomously and beats proprietary workflow agents.