🤖 Claude Opus 4.8 tops GPT-5.5 across all benchmarks 🤖
New rollout delivers 3x cost savings and parallel coding agents.

Anthropic launched Claude Opus 4.8, its most capable model yet, surpassing GPT-5.5 and Gemini 3.1 Pro across major benchmarks including agentic coding, computer use, financial analysis, and Humanity's Last Exam. Priced the same as Opus 4.7, the model is described as the least lazy and most honest in Anthropic's lineup, with a greater tendency to flag uncertainties. A new Fast mode delivers 3x cost savings, while claude.ai gains adjustable effort controls. Claude Code also receives parallel sub-agents through dynamic workflows. Anthropic teased a forthcoming “Mythos-class” AI system, intensifying competitive pressure among frontier labs heading into summer.

🧠 Demis Hassabis predicts AGI by 2030 🧠
DeepMind's CEO outlines AGI's rollout timeline and remaining capability gaps.

In a sit-down interview at Google I/O, Demis Hassabis shared his conviction that AGI will arrive by 2030, give or take a year. The Google DeepMind CEO identified the remaining gaps — world physics, memory, consistency, and continual learning — while expressing growing confidence the field is on track. He pinpointed oncology and immunology as the first diseases AI will crack and argued humans will retain an edge in taste, original thinking, and emotional connection. Post-AGI, Hassabis plans to pivot toward studying the nature of reality using AI — a timeline that signals how the industry's most powerful lab is allocating resources.

⛪ Pope Leo XIV's 42,000-word encyclical targets AI giants ⛪
The Vatican's sweeping regulation call demands AI governance accountability from corporations and governments alike.

Pope Leo XIV published Magnifica Humanitas, a 42,000-word encyclical addressed to the Catholic Church's 1.4 billion members, warning that AI must not be controlled by a handful of powerful companies. The document argues that private, transnational corporations already surpass many governments in capacity, and that lethal decisions must never be delegated to machines. The Pope called for robust legal frameworks, independent oversight, and political accountability, while warning against addictive algorithms and deepfakes. Anthropic's Christopher Olah attended the Vatican presentation and echoed concerns about misaligned incentives inside AI labs. The encyclical represents one of the most significant institutional statements on AI governance to date.

INTERESTING TO KNOW

🧬 Biohub open-sources $500M protein biology world model 🧬

Biohub, the nonprofit research center backed by Mark Zuckerberg and Priscilla Chan's CZI, publicly released a comprehensive suite of AI-powered protein biology tools. The release includes ESMC, a language model trained on 2.8 billion protein sequences; ESMFold2, claiming hit rates of 36–88% on cancer and immune targets — outperforming AlphaFold; and ESM Atlas, covering 6.8 billion protein sequences and 1.1 billion predicted structures. All three are freely available, part of a $500 million Virtual Biology Initiative that could reshape drug discovery dynamics worldwide.

🧮 AlphaProof Nexus solves 9 open Erdős problems 🧮

Google DeepMind's AlphaProof Nexus solved 9 open Erdős mathematics problems — including two unsolved for 56 years — along with 44 open OEIS conjectures. The system pairs AI models with the Lean proof assistant, which verifies each logical step to eliminate hallucinations. Some solutions reportedly cost only a few hundred dollars to generate. The announcement landed just one day after OpenAI claimed its own single Erdős result, sharpening the rivalry between the two labs in formal mathematical reasoning.