🎨 ChatGPT Images 2.5 ships with 50% less latency 🎨
OpenAI’s speed-focused rollout cuts image generation time in half across ChatGPT.

OpenAI released ChatGPT Images 2.5, a major upgrade to its image generation stack featuring sharper details, better reference-image preservation, and up to 50% lower latency compared to Images 2.0. The update introduces Sketch, a new mode that converts rough doodles into polished images via an @Sketch prompt, alongside comments for granular edits, reusable templates, and shareable prompts. Under the hood, two new models — Sunburst and Flare — power the upgrade and currently hold the #1 and #2 spots on Arena AI’s Image leaderboards. Both are accessible via the API. The release tightens OpenAI’s grip on AI-powered creative tools, a space seeing fierce competitive pressure from Google and Midjourney.

🧬 DeepMind opens AlphaGenome Atlas for free 🧬
Free access to a 1-petabyte genomic database could reshape precision medicine.

Google DeepMind launched AlphaGenome Atlas, a freely searchable 1-petabyte database that predicts the regulatory effects of all 9 billion possible single-nucleotide variants in the human genome. The tool gives scientists and clinicians an AI-powered resource to understand how individual DNA mutations may influence gene regulation and disease risk, which could accelerate diagnostics, drug discovery, and personalized medicine. Building on the legacy of AlphaFold, the atlas marks another landmark open-access contribution from Google’s AI research arm. By making this resource freely available, DeepMind positions itself at the intersection of AI and life sciences — a distribution advantage few competitors can match.

🤖 Meta ships Muse AI agent at $20/month 🤖
Meta’s pricing tiers signal an aggressive push into consumer AI agent subscriptions.

Meta unveiled Muse, billed as the world’s first personal AI agent, powered by a model called Muse Spark. Muse automates everyday tasks — booking travel, sending emails, managing Spotify playlists, shopping — all running on its own cloud computer in the background. It integrates with Gmail, Ticketmaster, and OpenTable, and can even code its own connections for unsupported apps. Security runs through Muse Secure VM with a dedicated “Sentinel” agent overseeing actions, Stripe-enabled payments, and built-in approval flows. Rollout begins in the U.S. with limited free usage and subscription tiers of $20 or $100/month across iOS, Android, WhatsApp, and muse.ai. The move marks Meta’s clearest bet yet beyond social media into consumer AI agents.

INTERESTING TO KNOW

🔬 OpenAI claims AI cracked Navier-Stokes $1M problem 🔬

OpenAI announced that an internal AI system, reportedly faster and more capable than GPT-6 Astra, produced a proof resolving the Navier-Stokes existence and smoothness problem — one of mathematics' seven Millennium Prize Problems carrying a $1 million bounty. The company says it ran roughly 10,000 agents simultaneously for 88 hours. However, the claim faces immediate controversy: NYU's Tristan Buckmaster and Anthropic's Levent Alpöge allege OpenAI's proof uses the same unusual method they had been developing and uploaded to Codex. The full proof has not been publicly released, leaving independent regulation and verification pending.

⚠️ Anthropic researcher quits citing 10% extinction odds ⚠️

The speed of frontier AI development prompted Anthropic researcher Jacob Coxon to resign after three years at OpenAI and Anthropic, publicly accusing both companies of "racing straight to self-improving superintelligence and gambling with our lives." Anthropic's own Alignment Science Lead, Evan Hubinger, backed Coxon's concerns, placing his personal odds of AI causing human extinction at 10%+ within the decade. The resignation adds to a growing pattern of high-profile safety departures and intensifies debate around governance of frontier AI, a clear adoption signal that internal alarm bells are ringing louder.