1. Anthropic Launches Computer Use for Claude
Anthropic has introduced a new computer use feature for Claude, allowing the model to control a user's mouse, keyboard, and screen. The capability is currently rolling out in research preview on macOS for Pro and Max subscribers via the Claude Cowork and Claude Code desktop apps. It can also be paired with the new Dispatch tool for remote mobile control.
2. OpenHarness Framework for Autonomous Agents
OpenHarness is a new release providing building blocks to create general-purpose AI agents in code. The framework supports the AGENTS.md specification, connects to Model Context Protocol servers, and allows agents to autonomously delegate work to other agents.
3. NVIDIA Kimodo Generates 3D Animations from Text
NVIDIA has released Kimodo, a tool that generates 3D character animations from text prompts in seconds. Trained on 700 hours of professional motion capture data, the system works on both human and robot skeletons and is available for free use on Hugging Face.
4. MiniMax M2.7 Offers 90% of Claude Opus Quality at 7% Cost
Recent testing compares MiniMax M2.7 against Claude Opus 4.6, showing the MiniMax model delivers 90% of Claude's quality at 7% of the cost. While MiniMax excels in bug detection and floating-point calculations, Claude remains more thorough in providing comprehensive fixes and vulnerability solutions.
5. Claude Code Adds Cloud-Based Recurring Tasks
Claude Code now supports scheduling recurring cloud-based tasks. This update allows Claude to run scheduled operations via cloud infrastructure without requiring users to keep the local desktop application running.
6. Luma AI Launches Uni-1 Image Generation Model
Luma AI has launched Uni-1, a new image generation model that reportedly outscores competitors from Google and OpenAI. The startup claims the new model achieves this performance while costing up to 30 percent less to operate.
7. MiniMax Publishes AI Coding Agent Skills Repository
MiniMax has published a repository of development skills designed to be plugged into AI coding agents. The repository provides structured guidance for frontend, full-stack, Android, iOS, and shader development, with installation instructions for tools like Claude Code, Cursor, and Codex.
8. Mozilla.ai Introduces Cq for Agent Learning
Mozilla.ai has introduced Cq, a project aiming to create a shared learning standard for AI coding agents. Functioning conceptually like a Stack Overflow for agents, Cq allows any model or agent to propose 'knowledge units' to improve collective agent capabilities.
9. Alibaba Releases LumosX for Video Generation
Alibaba has released LumosX on Hugging Face, an ICLR 2026 framework for personalized multi-subject video generation. The system uses relational attention mechanisms to associate specific identities with their corresponding attributes.
10. Pi Agent Now Available on Hugging Face
The Pi agent is now available directly on Hugging Face for compatible MLX models. Developers can access and deploy the agent straight from the platform's 'Use this model' menu.
11. Hugging Face Adds Batch Jobs for Image Enhancement
Hugging Face Pro users can now run batch image enhancement tasks using Hugging Face Jobs. The feature allows developers to scale image processing workflows directly on the platform's infrastructure.
12. Astrolabe Paper Explores RL for Video Models
Researchers have published 'Astrolabe', a new paper detailing methods for steering forward-process reinforcement learning. The research specifically targets improvements for distilled autoregressive video models.
13. HopChain Introduces Multi-Hop Data Synthesis
A new paper titled 'HopChain' introduces a multi-hop data synthesis approach. The method is designed to improve generalizable vision-language reasoning in AI models.