DeepSeek released preview versions of V4 today, featuring a 1.6-trillion-parameter Pro model and a leaner 284B Flash variant, both with 1 million-token context windows. The company claims V4-Pro-Max outperforms OpenAI's GPT-5.2 and Gemini 3.0 Pro on multiple reasoning benchmarks — at roughly one-tenth the per-token price. DeepSeek highlighted a new "Hybrid Attention Architecture" for improved long-conversation memory, and the timing — exactly one year after the original DeepSeek shock — is clearly deliberate.
OpenAI's first full base-model retrain since GPT-4.5 launched yesterday and hit the API today. GPT-5.5 tops the Artificial Analysis Intelligence Index at 60 and scores 82.7% on Terminal-Bench 2.0, with significant gains in agentic coding and multi-tool orchestration. Priced at $5/$30 per million tokens (input/output), it ships to Plus, Pro, Business, and Enterprise tiers across ChatGPT and Codex. "What is really special about this model is how much more it can do with less guidance," said OpenAI President Greg Brockman.
Anthropic launched a new wave of app connectors today, linking Claude to Spotify, Uber, Uber Eats, Instacart, TurboTax, Credit Karma, TripAdvisor, Booking.com, AllTrails, Audible, Resy, StubHub, Taskrabbit, Thumbtack, and Viator. The move positions Claude as a front door for daily consumer tasks, not just a chatbot. Anthropic emphasized user data from connected apps won't be used for training.
China's two tech giants are competing to join DeepSeek's maiden funding round, with valuation expectations surging from $10B to over $20B as investor demand accelerates. Tencent has proposed acquiring up to 20% of the startup, though DeepSeek is resisting ceding that much control. No deal is finalized, but a close would mark a watershed moment for China's independent AI sector.
After weeks of user backlash, Anthropic confirmed today that three separate engineering errors degraded Claude Code's performance from March through mid-April: reasoning effort was silently reduced to "medium," a bug caused the model to discard its own reasoning history mid-session, and a system prompt capped responses at 25 words between tool calls. All three have been fixed as of April 20, and Anthropic reset usage limits for all subscribers on April 23.
University of Cambridge researchers engineered a nanoelectronic device using modified hafnium oxide that mimics how neurons process and store information simultaneously. Unlike conventional chips that waste energy shuttling data between memory and processing units, the device operates at ultra-low power, addressing one of AI's biggest scaling bottlenecks: electricity consumption.
Elon Musk confirmed that the $20-25B Terafab semiconductor facility at Giga Texas will use Intel's 14A fabrication technology, with Tesla operating the pilot line and SpaceX handling high-volume manufacturing. The deal makes Terafab Intel's first confirmed anchor customer for the 14A node, giving Intel's struggling foundry business a lifeline while Musk vertically integrates AI hardware across Tesla, SpaceX, and xAI.
Speaking at the Digital World Conference in Geneva, Nobel laureate Geoffrey Hinton pushed back against industry arguments that regulation slows progress, arguing the real danger is building powerful systems with no governance at all. He estimated "maybe one percent" of AI work is going toward safety, and warned it remains unclear whether humanity can coexist with superintelligent AI.
State legislatures are racing ahead of Congress on AI regulation. Nineteen new AI bills have been signed into law this month, with Utah's governor signing nine alone. Connecticut passed a comprehensive AI bill regulating frontier model developers and creating a state AI sandbox. Meanwhile, the White House released a national AI legislative framework calling for federal preemption of certain state laws — setting up a federal-state collision course.
Adobe rebranded Experience Cloud as CX Enterprise, an agentic AI platform featuring persistent AI agents called "Coworkers" that orchestrate marketing and CX workflows across systems. The Adobe Marketing Agent is now GA inside Microsoft 365 Copilot and entering beta in Claude Enterprise, ChatGPT Enterprise, Gemini Enterprise, and IBM watsonx. It's a clear signal that enterprise software is shifting from tool-centric to agent-centric architecture.
