Models & Research

Mistral renames Le Chat to Vibe and turns it into a full agent that reads your email, writes reports, refactors code, and ships pull requests. One app for everything.

OpenAI polishes GPT-5.5 Instant for more readable answers but pulls the Canvas panel from its new models. At the same time, o3 and GPT-4.5 are being phased out of ChatGPT.

An internal OpenAI reasoning model autonomously disproved the Erdős unit distance conjecture with a 125-page proof. Mathematicians are calling it a milestone.

Anthropic's co-founder warns of existential AI risks at Oxford. Meanwhile, his company markets Claude Code more aggressively than ever. Time calls it 'A Tale of Two Anthropics'.

Deleted API keys stay active for 23 minutes, developers get five-figure bills from unauthorized Gemini calls, and shadow agents undermine security policies. The AI security landscape is sobering.

Simon Willison has released Datasette Agent — an extensible AI assistant that translates natural language into SQL and makes databases explorable.

Mistral AI acquires Austrian startup Emmi AI, specialists in physics simulation and digital twins. The goal: AI-powered real-time simulations for aerospace, automotive, and semiconductors.

Google went all-in at I/O 2026: Gemini 3.5 Flash, a personal AI agent called Spark, and a new creative model. The competition for Claude just got fiercer.

Chromebooks are yesterday's news. Google introduces Googlebook — an AI-native laptop with 'Magic Pointer' and vibe-coded widgets.

While OpenAI opens GPT-5.5-Cyber to European authorities, the EU is still waiting for access to Anthropic's Mythos. Two strategies, one dilemma.

Tomorrow Google previews Android 17 at the Android Show. The headline: Gemini AI becomes part of the operating system — no longer just an app.

Google I/O kicks off May 19. Expected: Gemini 4, a new desktop OS called Aluminium, and Google's answer to Claude Code. Here's what's coming.

The Trump administration is considering an executive order that would require AI models to undergo testing before release — like drugs at the FDA.

Google is developing Remy, a personal AI agent for Gemini. Currently being tested by Google employees internally, it could be unveiled at I/O in May.

OpenAI replaces GPT-5.3 Instant with GPT-5.5 Instant as the default. Fewer hallucinations, fewer words, more personalization — plus access to old chats as memory.

Google is redesigning the Gemini app from scratch. New animations, new input field, new philosophy. And on iOS, it looks particularly sharp.

After years of silence, Google is launching new smart speakers with Gemini — and opening the platform to third parties like Walmart.

Mistral merges chat, reasoning, and coding into a single 128-billion-parameter model. Plus cloud agents for Vibe and a new work mode for Le Chat.

First slam Anthropic for restricting Mythos, then restrict GPT-5.5-Cyber behind the same kind of velvet rope. OpenAI just reinvented the gatekeeping it recently ridiculed.

BioMysteryBench tests AI on real bioinformatics puzzles. Claude Mythos solves 30% of problems that stump human experts.