Top stories in artificial intelligence — last 24 hours
Google is investing up to $40 billion in Anthropic in the largest AI bet by a Big Tech company to date. The deal is structured as $10 billion at a $350 billion valuation — committed now — with an additional $30 billion contingent on Anthropic hitting performance milestones. Beyond cash, Google Cloud is committing five gigawatts of computing capacity to Anthropic over a five-year window, with scope for further expansion. The announcement follows Amazon's pledge of up to $25 billion and pushes Google's cumulative stake past $13 billion. Anthropic's annualised revenue has surpassed $30 billion — up from roughly $9 billion at end-2025 — driven by explosive growth in coding-focused enterprise contracts.
SpaceX (now merged with xAI) has struck a deal giving it the right to acquire AI coding startup Anysphere (Cursor) for $60 billion later this year, contingent on the SpaceX IPO this summer. As an alternative, SpaceX can pay Cursor $10 billion for their joint work and walk away. Cursor was valued at $29.3 billion in November 2025 after closing a $2.3 billion Series D — making the $60 billion price a swift doubling. The deal is framed as a response to xAI lagging behind Anthropic and OpenAI in coding tools. Separately, CNBC reported that Microsoft had also evaluated a potential acquisition but did not proceed.
OpenAI released GPT-5.5 and GPT-5.5 Pro for paid ChatGPT subscribers (Plus, Pro, Business, Enterprise) and API customers, just six weeks after GPT-5.4. The company says the model matches GPT-5.4's per-token latency while running at "much higher intelligence" and completing tasks in fewer tokens — with the sharpest gains in agentic coding, computer use, and early scientific research. API pricing: $5/1M input tokens and $30/1M output for the standard tier; $30/$180 for Pro. Context window is 1 million tokens. GPT-5.5 also powers Codex and is rolling out to OpenAI's enterprise offerings.
Anthropic released Claude Opus 4.7, the first Claude model with high-resolution image support — maximum resolution jumps to 2,576 px / 3.75 MP. The model also shows improved performance on long-running software engineering tasks. Alongside the model, Anthropic launched Claude Design, an Anthropic Labs product that lets users collaborate with Claude to produce visual outputs: designs, prototypes, slides, and one-pagers. Memory for Claude Managed Agents entered public beta under the new managed-agents-2026-04-01 API header, enabling persistent agent memory across sessions.
Google unveiled two new research agents in public preview via the paid Gemini API. Deep Research is optimised for speed and low latency — ideal for real-time interactive surfaces. Deep Research Max applies extended test-time compute for maximum comprehensiveness, designed for async background workflows like nightly jobs. Both agents can fuse open-web data with private enterprise information through a single API call, generate native charts and infographics inside reports, and connect to arbitrary third-party data sources via Model Context Protocol (MCP). Separately, Google launched the Gemini Enterprise Agent Platform — an evolution of Vertex AI offering unique agent identities and governance tooling.
xAI's Grok Computer is now in targeted private beta, with Elon Musk confirming large-scale public testing is "launching within days." The agent takes full control of a desktop — operating apps, clicking buttons, typing text, filling forms, and executing complex multi-step workflows without API access, relying instead on screen-pixel reading. Grok Computer is part of the broader Macrohard initiative — a joint Tesla-xAI venture backed by a $2 billion Tesla investment — that Musk framed as a deliberate jab at Microsoft. Alongside this, xAI quietly released Grok 4.3 Beta on April 17 for SuperGrok Heavy subscribers.
Cursor's major version 3 release reframes the IDE as an agent management platform rather than a code editor. Key additions: real-time reinforcement learning for Composer that trains and deploys improved checkpoints as often as every five hours using live user interactions; Canvases — interactive React-based visual interfaces agents can generate for data exploration, PR reviews, and eval analysis; and an improved Agents Window with tiled layout for running multiple agents in parallel, more reliable voice dictation, and direct diff-to-file navigation. Cursor also launched agent-based coding workflows targeting teams.
Anthropic opened a limited preview of Claude Mythos to 11 companies and organisations focused on finding and fixing cybersecurity vulnerabilities. The model is capable of identifying critical vulnerabilities in major operating systems and web browsers. In independent evaluation, the UK AI Security Institute reported that Mythos solved a 32-step cyber range challenge called "The Last Ones" in 3 out of 10 attempts — the first AI model to complete it end-to-end. The controlled rollout reflects Anthropic's safety-first approach to releasing models with advanced offensive capabilities.