OpenAI rolled out a major Agents SDK update adding native sandbox execution, long-horizon agent harness, and multi-agent handoffs. Enterprise teams can now run agents in isolated compute environments with full provider flexibility across 100+ LLMs.
Python-first, TypeScript support coming.
Stanford HAI's 2026 AI Index is out: China has erased the US lead on AI benchmarks, trading places at the top. Generative AI hit 53% global adoption — faster than any tech in history.
Transparency drops: 90%+ of notable models now come from private firms that no longer disclose training details.
Meta launches Muse Spark, its first proprietary AI model from new Meta Superintelligence Labs. Natively multimodal with tool use & visual chain-of-thought.
Major shift: no public weights unlike Llama — API access is invite-only. Scores 58% on Humanity's Last Exam.
Available on meta.ai & Meta AI app now.
🔐 Anthropic now requires government ID + live selfie to access certain Claude features. Verified via Persona Identities — data not used for model training.
A first among major AI chatbots, raising privacy debate: should you hand your passport to an AI company?
Anthropic is preparing to launch Claude Opus 4.7 alongside a new AI design tool for websites & presentations. The release also includes a unified Claude Code interface for multi-project management and a beta Claude for Microsoft Word integration.
Launch expected imminently.
Google launched Gemini 3.1 Flash TTS, a next-gen text-to-speech model supporting 70+ languages and 200+ audio tags for granular speech control. It scores 1,211 Elo on TTS leaderboard, supports native multi-speaker dialogue, and all output is watermarked with SynthID.
Now live on AI Studio & Vertex AI.
Anthropic is preparing to release Claude Opus 4.7, its next flagship AI model, potentially as early as this week. The upgrade builds on Opus 4.6's coding and task-execution strengths, and will launch alongside a new AI design tool capable of building websites and presentations.
Stronger safety controls and improved alignment measures are also expected.
Google DeepMind released Gemini Robotics-ER 1.6, a major upgrade for autonomous robots. Key improvements: spatial reasoning, multi-view task detection, and 93% accuracy in reading industrial gauges.
Boston Dynamics integrated it into Spot for facility inspections. Now available via Gemini API and AI Studio.
Google launched AI Skills for Chrome — save your best Gemini prompts as one-click reusable tools. Trigger with '/' or '+' across any tab.
Pre-built skill library covers productivity, shopping, recipes & more. Rolling out to English-US users on Mac, Windows & ChromeOS.
OpenAI launched GPT-5.4-Cyber, a cyber-permissive variant of GPT-5.4 built for defensive security. Key capability: binary reverse engineering of compiled software without source code.
Access limited to vetted researchers & vendors via Trusted Access for Cyber program.
Anthropic redesigned Claude Code desktop for parallel agents. New features: multi-session sidebar, drag-and-drop layout, built-in terminal & file editor, side chat (⌘+;), and SSH support on macOS.
Parallel agent workflows just got a serious upgrade.
Google Home rolls out new Gemini voice upgrades: smarter playlist recognition (even with background noise), advanced notes/lists management (e.g. "remove all vegetables from my shopping list"), and parental controls with screen time limits.
Updates available via the Google Home app.
Google's AI Mode now books restaurants in 8 new countries — UK, Australia, Canada, India, Singapore, Hong Kong, New Zealand & South Africa. The agentic feature interprets natural language requests, checks real-time availability, and connects directly to local booking platforms like OpenTable & TheFork.
Search just became action. 🍽️
OpenAI has acquired Hiro Finance, an AI-powered personal finance startup backed by Ribbit and General Catalyst. Founder Ethan Bloch (previously created Digit, sold for $200M) and ~10 team members join OpenAI to build financial planning features into ChatGPT.
Hiro shuts down April 20. https://techcrunch.com/2026/04/13/openai-has-bought-ai-personal-finance-startup-hiro/
Stanford HAI's 2026 AI Index Report reveals China has erased the US lead in AI model performance — top models now separated by just 2.7%. GenAI reached 53% global adoption in 3 years, faster than PCs or the internet.
US AI private investment hit $285.9B in 2025 (23x China), yet public trust in AI regulation fell to 31% in the US. Over 90% of notable models now come from private companies.
Microsoft is quietly building its own OpenClaw-like autonomous AI agent for M365 Copilot. Led by newly hired Omar Shahine, the team aims to move Copilot beyond chatbots to agents that handle tasks end-to-end.
An early preview may land at Build 2026 in June.
DeepSeek V4 launch is imminent — expected late April 2026. The 1-trillion-parameter model features a 1M-token context window and runs on Huawei Ascend 950PR chips, marking the first frontier AI built on Chinese semiconductor infrastructure.
Apache 2.0 license confirmed. Alibaba, ByteDance & Tencent already queuing for cloud deployment.
Meta is building photorealistic AI-powered 3D characters — including a digital version of Zuckerberg that employees can interact with. Zuckerberg personally trained and tested his own AI clone, which can offer feedback and conversation in his place.
Meta is betting big on personal AI superintelligence to catch up with OpenAI and Google.
Anthropic debuts Claude for Word in beta — a native sidebar add-in for Microsoft Word that delivers AI-powered editing, clickable citations, and tracked changes. Available for Team & Enterprise users on Mac and Windows.
Legal contract review is the first featured use case.
Meta launched Muse Spark, its first fully proprietary AI model — abandoning the open-source Llama legacy. Under Alexandr Wang's Meta Superintelligence Labs, Muse Spark ranks 4th globally on AI benchmarks using 10x less compute than Llama 4.
A major strategic pivot for Meta's 3B-user ecosystem.
