GPT-5.4 scored 74.0% on ARC-AGI-2 👀

GPT-5.4 Pro got 83.3%, getting close to a level of Gemini 3 Deep Think.

Gemini 3.1 Pro is now available on Stitch 👀

OpenAI releases GPT-5.4 model on ChatGPT and Codex

OpenAI has released GPT-5.4, advancing reasoning, coding, multimodal understanding, and agent-based workflows. Available in ChatGPT, API, and Codex, it supports longer contexts, complex problem solving, tool use, and enterprise-scale automation.

🗞 #chatgpt

ICYMI: Atlas browser from OpenAI now supports saved prompts. Users can now "bookmark" any prompt to use it later in the Atlas browser.

Saved prompts can utilize tools as well.

Use @ or / to invoke 👀

Google tests Live Voice Mode and Design Studio on Stitch

Google’s Stitch is set to expand with voice-driven UI generation, design system imports from URLs, advanced prototyping controls, and a new Studio tab for audits, marketing kits, and developer specs, positioning it as a comprehensive end-to-end product design platform.

🗞 #stitch

BREAKING 🚨: Google is working on Live Voice Mode for Stitch! Users will be able to trigger design changes, discuss potential improvements, or have a live design critique.

Besides that, loads of new generations will be available soon as a part of a new design "Studio" feature.

Cursor debuts Automations for scheduled coding agents

Cursor has launched Automations, enabling always-on coding agents that run on schedules or external triggers like Slack, GitHub, and PagerDuty. Built on cloud sandboxes, it targets review, incident response, and maintenance, expanding Cursor into a workflow layer for engineering teams.

🗞 #cursor

Luma unveils UNI-1, unified reasoning image model

Luma has launched UNI-1, a unified model that combines visual understanding and image generation in one system. It performs structured reasoning during synthesis and supports advanced editing and control.

🗞 #ai

Microsoft tests AI-generated Discover feed on Copilot web

Microsoft is testing an AI-generated Discover page in Copilot that builds a personalized feed from user history and memories. Updated daily, it replaces curated news with tailored text and images, signaling a shift toward generative UI despite limited customization options.

🗞 #microsoftcopilot

Updated Clawd animation 🦀

ICYMI: Scheduled tasks are now fully available on the desktop app.

Users can now also prevent computer from sleeping when the Claude app is open, so scheduled tasks can run.

Open Clawd 👀

Perplexity is officially rolling out "Skills" for Perplexity Computer to let users create and execute more complex and reusable workflows!

Skills are new Programs 👀

Vellum Labs launches waitlist for private on-device AI assistant

Vellum Labs has launched a public waitlist for Vellum, a personal AI assistant that runs entirely on a user’s Mac with no cloud processing or data collection. It handles tasks like email drafting, scheduling, calls, and shopping while keeping all data on-device.

🗞 #sponsored

OpenAI launched Codex Security, a new application security agent that can find and fix security vulnerabilities.

"Codex Security leverages OpenAI’s frontier models and the Codex agent"