AI news digest — April 21, 2026
10 items, each with its source.
OpenAI Announces GPT-5 General Availability and API Pricing
OpenAI has officially launched GPT-5 for all developers, introducing a new tiered API pricing model. The model boasts significant improvements in reasoning and context window size, supporting up to 2 million tokens.
openai.comEuropean Union Enforces AI Act Compliance Deadline
The European Union has officially begun enforcing the strict compliance measures outlined in the AI Act. Companies failing to meet the transparency and risk assessment requirements now face fines of up to 7% of their global revenue.
reuters.comDeepMind Publishes Breakthrough in Room-Temperature Superconductor AI
Google DeepMind researchers published a paper detailing an AI model that successfully predicted stable room-temperature superconductor materials. This breakthrough could revolutionize energy transmission and quantum computing hardware.
arxiv.orgNVIDIA Unveils R100 Tensor Core GPU Architecture
NVIDIA has revealed its next-generation R100 GPU architecture, promising a 5x increase in training efficiency for trillion-parameter models. The new chips utilize advanced optical interconnects to bypass traditional bandwidth bottlenecks.
anandtech.comAnthropic Secures $4 Billion Series E Funding Round
AI safety startup Anthropic has closed a massive $4 billion Series E funding round led by major sovereign wealth funds. The capital will be used to expand their constitutional AI research and build next-generation data centers.
bloomberg.comApple Integrates On-Device LLM into iOS 20 Core
Apple announced the deep integration of its proprietary 'Apple Intelligence' LLM directly into the iOS 20 kernel. This allows for zero-latency, offline natural language processing for all native applications while preserving user privacy.
techcrunch.comBoston Dynamics Unveils Fully Autonomous Construction Robot
Boston Dynamics introduced 'BuildBot', a fully autonomous humanoid robot designed specifically for complex construction tasks. The robot uses advanced spatial AI to navigate active sites and assemble scaffolding without human intervention.
theverge.comUN Establishes Global AI Watermark Standard Agreement
The United Nations has ratified a global agreement requiring all member states to implement standardized cryptographic watermarks for AI-generated media. The initiative aims to combat the rising tide of deepfakes in global elections.
news.un.orgMIT Researchers Solve LLM Hallucination via Fact-Checking Layer
A new paper from MIT demonstrates a novel neural architecture that significantly reduces factual hallucinations in LLMs. The system uses a deterministic, real-time fact-checking layer that cross-references outputs against a verified knowledge graph.
news.mit.eduNeuralink Competitor 'Cerebra' Exits Stealth with $200M
Brain-computer interface startup Cerebra has emerged from stealth mode, announcing a $200 million seed round. The company claims its non-invasive neural headband can interface directly with consumer LLMs for thought-to-text generation.
techcrunch.comFeed, daily deep-dive and bytes — readable offline, with push alerts for the topics you follow.