AI News Brief | 2026-05-24

AI Industry Overview

The AI industry is undergoing a seismic shift toward agentic systems. Google unveiled Gemini 3.5, a "frontier intelligence with action" model that powers a more proactive Gemini app offering 24/7 autonomous help. This marks a major move beyond traditional chat toward continuous, task-completing agents. (Gemini 3.5 blog, Gemini app blog)

Anthropic is on the verge of closing a $300+ billion funding round that could value the company over $900 billion, surpassing OpenAI as the world's most valuable AI startup. Strong Q2 revenue growth (estimated $109B) points to its first profitable quarter. The company also launched Claude for Small Business — AI agents that integrate with existing tools like Google Workspace and Slack. (Bloomberg, Quasa)

In China, Alibaba's latest AI model ran autonomously for 35 hours to optimize code for its own custom chip, demonstrating self-improving AI systems. Meanwhile, DeepSeek made its 75% API discount permanent, pricing output tokens at least 34x below GPT-5.5 — a massive price war escalation. (The Decoder)

OpenAI announced it is shutting down Sora, its AI video generation model, app, and API — a surprise move that leaves a gap that Google aims to fill with Gemini Omni. (VentureBeat)

NVIDIA CEO Jensen Huang predicted AI infrastructure annual spending could rise from $1 trillion to $3–4 trillion, far above Wall Street expectations. The company's Q1 FY2027 revenue hit $816B, up 85% YoY. (IT之家)

The White House approved $9 billion for U.S. spy agencies to accelerate AI adoption. In contrast, Microsoft released a report finding that deploying AI in certain work scenarios currently costs more than paying human wages. (The Inquirer, Fortune)

Google CEO Sundar Pichai redefined links as just "a part" of search, signaling a deeper integration of AI-generated answers. Mistral AI acquired physics simulation startup Emmi AI to strengthen industrial digital twins. Across startups, Kakuna offers automated codebase hardening, and the feishu-claude-code-bridge open-source project connects Feishu with Claude Code CLI. (The Decoder, Mistral AI, X/@dotey, X/@swyx)

---

Gemini

Google announced Gemini 3.5 with "action" capabilities — the model can autonomously execute multi-step tasks across apps and services. The Gemini app is being updated to become a proactive, always-on assistant that anticipates user needs. This redefines the internet search experience as agents take over routine online work. (Gemini app blog, ynetnews)

Claude & Anthropic

Anthropic is reportedly finalizing a $300B+ funding round, pushing its valuation past $900B and overtaking OpenAI. The company launched Claude for Small Business, embedding AI agents into tools like Gmail, Slack, and Office 365. A new Claude Mythos Preview finds software bugs faster than developers can patch them, raising both promise and alarm. Anthropic also struck a compute deal with SpaceX and increased extended usage limits for Claude. (Anthropic blog, The Decoder)

DeepSeek

DeepSeek permanently applied a 75% discount to its flagship model API. Output tokens now cost at least 34x less than GPT-5.5. Founder Liang Wenfeng publicly declared an AGI goal, while a $9.6B (70B yuan) funding round advances. (Bloomberg, The Business Times)

Sora

OpenAI is discontinuing Sora (video generation model, app, and API). Users are being notified of the shutdown timeline. Google quickly announced Gemini Omni to fill the gap, expanding AI creation tools. (OpenAI Help Center, IBTimes)

Cursor

Cursor launched Composer 2.5, a model for long-running AI coding tasks at a cheaper cost. In a blockbuster rumor, SpaceX plans to acquire Cursor for $60 billion after its record IPO. Cursor also improved its Automations feature. (The Indian Express, The Next Web)

Qwen / Alibaba

Alibaba's Qwen model ran autonomously for 35 hours to optimize code for the company's custom chip, showcasing self-directed AI engineering. The release of Qwen3.7-Max emphasizes an "agent-first" architecture for coding tasks. (The Decoder, Alibaba Cloud)

ChatGPT

OpenAI introduced invisible AI markers in ChatGPT-generated images that anyone can detect — a major transparency push. A new ChatGPT PowerPoint Beta brings AI inside presentations. The voice mode now supports filling out forms: upload a document and speak the answers. (TechTimes, TechMyMoney, X@ChatGPTapp)

Claude Code

Claude Code v2.1.149 added /usage command categories, keyboard scrolling for diffs, and enterprise settings. Auto mode is now available on the Pro plan and supports Sonnet 4.6 and Opus 4.7. Co-creator Sid Bidasaria discussed expanding Claude Code beyond engineers to knowledge workers. (GitHub Releases, X@ClaudeDevs, Moneycontrol)

Devin

Cognition broadened Devin with Windows support, testing upgrades, and incident management. Auto-Triage now monitors and fixes production incidents automatically. Cognition also extended Devin's capabilities to Android app development. (TipRanks, Headsup AI)

Bolt.new

Bolt expanded into Microsoft Azure and M365 for enterprise AI development, making it easier to build and deploy apps on Microsoft's cloud. Bolt.new also released new release notes and a discussion on how vibe-coded prototypes need design systems. (Create With, Bolt.new support, Worktechjournal)

Doubao

ByteDance's Doubao introduced a "scan to pay" feature with subscription plans up to 500 yuan/month. The chatbot is also being sued for misleading ticket refund fee information and failing to deliver promised compensation. (Caixin Global, Gate News)

ElevenLabs

ElevenLabs partnered with Splice for next-gen AI music creation tools and with Spotify for an AI-powered audiobook publishing tool that turns text into narrated audio. (MusicTech, TimeBulletin)

Perplexity

Perplexity open-sourced Bumblebee, a read-only supply-chain scanner for developer endpoints. The Comet AI browser for iOS received eight major improvements, including better search and agent interactions. (MarkTechPost, 9to5Mac)

Kimi

Moonshot AI's Kimi WebBridge lets AI agents drive your browser while keeping your data local — a step toward truly autonomous web agents that can fill forms, navigate sites, and extract information without compromising privacy. (Decrypt)

Kling AI

Kling AI made its debut at the Cannes Film Market with a filmmaker initiative, showcasing how its video generation AI can serve animated features, Hollywood series, and experimental shorts. Global filmmakers are leveraging Kling to push storytelling boundaries. (GlobeNewsWire)

GitHub Copilot

GitHub Copilot now uses Auto model selection in VS Code that routes based on your task. Copilot Chat gained semantic issue search — find issues by describing them in natural language. GPT-5.3-Codex is now the base model for Copilot Business and Enterprise. (GitHub Changelog, GitHub Changelog, GitHub Changelog)

v0 by Vercel

Vercel introduced the new v0 — a redesigned coding agent that better handles complex frontend tasks. A companion blog details how Vercel made v0 an effective coding agent. (Vercel blog, Vercel blog)

Windsurf

Cognition (maker of Devin) acquired the remaining Windsurf team and technology, reconnecting with Anthropic in the process. Windsurf AI 2.3.9 was also released with various improvements. (xix.ai, warp2search)

Aider

Aider v0.24.0 was released. An XDA article discusses why a developer stopped forcing every coding job through Claude and started using Aider instead, praising its flexibility. (PyPI, XDA)

ChatGLM / Zhipu AI

Zhipu AI jointly launched GLM-5.1 with TileRT, achieving 400 tokens/s — a world record for inference speed. The company highlighted its journey from ChatGLM-6B open-source to ChatGLM-4 multimodal capabilities. (AIbase)

Cline

Cline introduced the Cline SDK — an upgraded agent runtime — and rebuilt Cline upon it. CLI versions 3.0.11 and 3.0.13 were released with stability improvements. (Cline blog, GitHub Releases)

Cody

CodeWords raised €7.6 million ($9M) in seed funding to expand its proactive AI business agent Cody across European SMEs. The agent works inside existing workflows to automate routine tasks. (Beinsure)

Midjourney

An analysis piece titled "Midjourney's TPU regret is a warning for AI startups" explores how Midjourney's decision to rely on Google TPUs may have backfired, offering lessons for hardware strategy and vendor lock-in. (Startup Fortune)

Notion AI

Notion turned its workspace into a hub for AI agents with the launch of the Notion Developer Platform (3.5). Developers can now create and deploy agents that read, write, and manage Notion data autonomously. (TechCrunch, Notion Releases)

Pika

Pika released an MCP (Model Context Protocol) integration that lets users "Pika-fy" different agents, giving them personality and style. The feature allows generating videos inside Claude using Pika agents. (Eyerys)

Flux

Flux launched a new steerable agent for hardware design, enabling engineers to describe requirements in natural language and get optimized circuit layouts and simulations. (Flux blog)

Consensus

Consensus AI raised $19.2M to expand its evidence-based AI search engine for academic research. The platform provides university access and is growing among researchers. (Awaira)

CopilotKit

CopilotKit is redefining the agentic AI stack in 2026 with its frontend-first approach for building copilot experiences. The company also launched an Enterprise Intelligence Platform as a persistence layer for agentic applications. (MarkTechPost, CopilotKit blog)

---

Other notable mentions: StepAudio 2.5 launched real-time speech with personality customization; Replit Agent integrated with Squidler for automated testing; NVIDIA's Nemotron-Labs diffusion language model aims for light-speed text generation; Google DeepMind's AlphaProof Nexus combines LLMs with Lean for formal verification of math proofs; and ViggleAI made motion capture and character animation easier.