What's new? Verdent a coding agent 76.1% pass@1 on SWE-bench is live as a Visual Studio Code extension plus standalone app; it runs plan-code-verify subagents for debug and review;

What's new? Marble now builds 3D worlds from text, images, videos or simple 3D layouts with diverse exports; Chisel lets users sculpt 3D assets and join existing models;

FlowithOS by flowith AI rethinks desktop browsing on Mac/Windows by merging web access with AI task logs in vertical tabs and a sidebar.

What's new? Code Arena assesses AI coding models over real-world dev cycles with agentic tool calls; it logs file operations with live coding and traceable IDs;

What's new? Poe by Quora now offers group chat, letting up to 200 users and multiple AI models collaborate in one conversation; The feature supports text, image, video, and audio models, including custom bots, and syncs chat history across all devices;

Grok 4.1 is now live for all users, offering faster, more creative responses and top leaderboard scores across grok.com, X, and mobile apps.

What's new? Anthropics launched Claude Opus 4.5 with long context; it adds a new API effort parameter and lower token pricing; expanded Claude for Excel and Claude for Chrome support.

Perplexity introduces a conversational AI shopping feature to US users, combining search, product discovery and checkout in one seamless chat experience.

What's new? Flux.2 is a family of image models available via API and openweight versions; it supports 10-image referencing, text rendering and 4mp editing;

We have released Qt Safe Renderer 2.2 Release candidate 1 for commercial license holders today. With this release candidate, you can use safety-critical items in Qt Quick Ultralite applications. You should use Qt Safe Renderer 2.2 Release candidate 1 together…

What's new? Runway released Gen-4.5 AI video model with cinematic quality on NVIDIA Hopper and Blackwell GPUs; new training and inference techniques support complex scenes;

What's new? Mistral AI launched Mistral 3 with compact and sparse mixture-of-experts models; all models are open source under Apache 2.0 and on major platforms;

What's new? Browsesafe-bench is a benchmark and detection model that secures browser-based AI agents; it simulates varied threat scenarios for research validation;