Morning Digest, April 25, 2026

April 25, 2026 · 14 newsletters · 5 overlapping stories

Top Stories

GPT-5.5 Launches as OpenAI’s Most Capable and Agentic Model Yet

(6 newsletters)

OpenAI released GPT-5.5 (codenamed “Spud”), its most capable model to date, topping benchmark scores across reasoning, agentic tasks, computer use, and coding. The model matches the speed of its predecessor while improving token efficiency and lands at $5/$30 per million input/output tokens, roughly half the cost of competing frontier coding models. It’s rolling out across paid ChatGPT plans and in Codex, and Nvidia has already deployed it to over 10,000 employees. After months of Anthropic’s Opus models leading leaderboards, sentiment is shifting back toward OpenAI, especially as Anthropic absorbed rate-limit and quality complaints in the same week.

Anthropic Publishes Claude Code Quality Postmortem

(5 newsletters)

Anthropic traced weeks of user complaints about degraded Claude Code quality to three separate accidental changes affecting Claude Code, the Claude Agent SDK, and Claude Cowork (the API was not affected). All issues were resolved on April 20. The company reset usage limits for affected subscribers and committed to stricter testing protocols, better code review, and tighter system prompt controls going forward. The postmortem drew significant attention from developer communities, with some viral posts about the regression generating millions of views before the fix landed.

Google Plans to Invest Up to $40 Billion in Anthropic

(3 newsletters)

Google is planning to invest up to $40 billion in Anthropic, with $10 billion upfront, to lock in compute capacity and stay competitive in the AI race. The investment follows Anthropic’s limited release of its latest model, Mythos, to a select group of partners. The news coincides with Anthropic’s valuation on secondary markets surpassing OpenAI’s $880 billion, hitting $1 trillion on Forge Global, driven by scarce available shares and growing demand for Claude Code and related tools.

Meta and Microsoft Announce Major Workforce Reductions

(3 newsletters)

Both companies announced significant workforce reductions tied to AI investment priorities. Meta will cut approximately 8,000 employees (10% of its workforce) beginning May 20 and leave roughly 6,000 open roles unfilled. Microsoft is offering voluntary buyouts to about 7% of its U.S. workforce, around 8,750 employees. Both are simultaneously plowing billions into AI infrastructure and data centers. The moves come ahead of April 29 earnings calls, signaling to Wall Street that management is willing to trade headcount for efficiency as compute budgets escalate. Amazon laid off roughly 30,000 workers over the past six months, and Oracle is also cutting, suggesting a sector-wide pattern rather than isolated restructuring.

DeepSeek Previews V4 Series with 1M-Token Context and Agent Focus

(3 newsletters)

DeepSeek previewed its next-generation flagship model in V4 Flash and V4 Pro variants, claiming top-tier performance in coding benchmarks with a 1-million-token context window and a new Hybrid Attention Architecture designed for long-context reasoning and agent-based tasks. The model runs natively on Huawei’s Ascend processors. Server capacity for V4 Pro remains limited due to a computing crunch, but pricing is expected to drop when Huawei’s Ascend 950 clusters come online later this year. Separately, Tencent and Alibaba are reportedly in talks to back DeepSeek at a $20 billion-plus valuation in its first-ever funding round.

Also Worth Knowing

Quick Hits