AI Daily Brief — April 24, 2026

OpenAI Apr 23 · Model Release

OpenAI ships GPT-5.5 — doubled price, 60% fewer hallucinations

OpenAI released GPT-5.5 on April 23, just six weeks after GPT-5.4. The model scores 88.7% on SWE-bench and claims 60% fewer hallucinations than its predecessor while matching GPT-5.4's per-token latency in real-world serving. Three variants ship: standard, Thinking (extended reasoning), and Pro (highest accuracy).

The release comes with a price hike: input doubles from $2.50 to $5 / 1M tokens, output from $15 to $30 / 1M tokens. GPT-5.5 Pro reaches $30/$180 per million tokens. Batch API pricing remains at 50% discount. The model is rolling out to Plus, Pro, Business, and Enterprise users in ChatGPT and Codex.

Source: openai.com · Yahoo Tech

Google Apr 22–23 · Enterprise

Google Cloud Next '26: Gemini Enterprise Agent Platform replaces Vertex AI

Google unveiled the Gemini Enterprise Agent Platform at Cloud Next '26, consolidating agent building, deployment, data integration, security, and optimization into a single offering that supersedes Vertex AI as its primary enterprise AI environment. Key additions include Agent Studio (low-code, natural language interface), an upgraded Agent Development Kit with a graph-based multi-agent orchestration framework, and new 8th-generation TPUs.

Google also announced a $750M fund to support agentic AI development in its partner ecosystem and a landmark deal with Merck — a multi-year investment valued at up to $1B to deploy an agentic platform across Merck's R&D, manufacturing, commercial, and corporate functions.

Source: cloud.google.com · The Next Web

Cursor Apr 21–22 · M&A

SpaceX secures $60B option to acquire Cursor, $10B coding AI project underway

SpaceX and AI coding startup Cursor have struck a deal giving SpaceX the option to acquire Cursor for $60 billion — or pay $10 billion for the companies' joint work on "coding and knowledge work" AI. The partnership pairs Cursor's AI expertise with SpaceX's Colossus supercomputer infrastructure.

SpaceX is delaying the formal acquisition until after its IPO this summer, preferring to avoid updating confidential financial filings before the listing and planning to use publicly-traded stock to finance the purchase. The deal follows SpaceX's February 2026 acquisition of xAI (Grok, the X platform) and underscores Musk's broader effort to build a full AI stack independent of OpenAI and Anthropic.

Source: TechCrunch · CNBC

Anthropic Security Apr 23–24 · Breach

Unauthorized group breaches access controls around Anthropic's Mythos cyber AI

An unauthorized group has reportedly gained access to Anthropic's closely guarded Claude Mythos Preview, the highly capable cybersecurity model that Anthropic described as too dangerous to release publicly when it launched on April 7. Anthropic is probing the incident, believed to have occurred through a third-party contractor — highlighting the difficulty of securing frontier systems when multiple vendors and access layers sit between the developer and deployment.

Former national cyber director Kemba Walden added to the concern in a Fortune interview, saying Mythos "can hack nearly anything" and that critical infrastructure operators are not yet prepared for AI-enabled adversaries at this capability level.

Source: CybersecurityNews · Fortune

Anthropic Apr 7 · Initiative

Project Glasswing: Anthropic uses Mythos to hunt zero-days across critical software

In response to Mythos's formidable offensive capabilities, Anthropic launched Project Glasswing — a defensive initiative giving major tech firms selective access to Mythos Preview to harden critical software. Partners include AWS, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, Microsoft, and Nvidia.

In initial runs, Mythos identified thousands of zero-day vulnerabilities across every major OS and web browser. One discovered bug had been dormant undetected for 27 years in OpenBSD. The findings are being shared broadly with the industry. Microsoft also confirmed it will integrate Claude Mythos Preview into its secure coding framework for threat detection and response.

Source: TechCrunch · Foreign Policy

Anthropic Apr 2026 · Developer Tool

Claude Code launches /ultrareview — multi-agent cloud code review, free until May 5

Anthropic shipped /ultrareview for Claude Code, a multi-agent cloud-based code review tool that analyzes branches or pull requests using a fleet of specialized agents in parallel. It's free for Pro and Max users through May 5. The tool can be invoked with /ultrareview on the current branch or /ultrareview <PR#> for a specific GitHub PR.

SemiAnalysis estimates Claude Code now accounts for roughly 4% of all public GitHub commits (as of March 2026), with projections suggesting 20% by year-end — underscoring the product's rapid adoption since launch.

Source: Anthropic · The New Stack

OpenAI Apr 16 · Developer Tool

OpenAI Codex gains desktop control, memory, 90+ plugins — 3M weekly active users

OpenAI's April 16 Codex overhaul dramatically expanded its agentic capabilities to rival Claude Code. The update adds computer use (Codex can see the screen and control mouse/keyboard), parallel subagents, persistent memory across sessions, image generation for mockups and designs, and over 90 plugins for external integrations.

The product surpassed 3 million weekly active users, up from 2 million just a month prior — making it one of the fastest-growing developer AI tools. The /best-of-n equivalent and Design Mode (from Cursor) remain differentiators, but Codex's plugin ecosystem is now its most distinctive edge.

Source: SiliconAngle · n1n.ai

Meta Microsoft Apr 24 · Industry

Meta and Microsoft cut thousands of jobs as AI investment surges

Meta announced it will lay off 10% of its workforce (~8,000 workers) beginning May 20, and will leave 6,000 open positions unfilled — all while simultaneously committing to $115–135B in AI capex for 2026, nearly double last year's spending. Microsoft is offering voluntary buyouts to roughly 7% of its US workforce (~8,750 employees).

Both companies frame the cuts as efficiency-driven to fund AI infrastructure. Meta's aggressive capex signals a determination to close the capability gap with OpenAI and Google. The announcements arrive as Big Tech collectively projects over $650 billion in combined AI infrastructure spending across 2026.

Source: FX Leaders · CNBC