Story cards stay below the sticky dock while audio, chapters, date, and brief navigation remain accessible.
OpenAI released GPT-5.5 (codename "Spud") on April 23, its first fully retrained base model since GPT-4.5. It ships with a 1M-token context window, scores 88.7% on SWE-bench, nearly doubles Claude Opus 4.7's FrontierMath score, and cuts hallucinations 60% versus GPT-5.4. API pricing doubles to $5/$30 per million input/output tokens — the steepest single-release jump in the GPT-5.x series.
DeepSeek released V4 Flash and V4 Pro, featuring a hybrid attention architecture with 1M-token context windows that rival GPT-5.5 on coding and reasoning benchmarks — trained for roughly $5M versus $100M+ for Western competitors. The release lands just as DeepSeek pursues its first outside funding round, putting the open-source pressure squarely on OpenAI's pricing narrative.
Anthropic is probing a breach of its most dangerous model, Claude Mythos Preview — a cybersecurity-focused system capable of autonomously discovering thousands of zero-day vulnerabilities. A Discord group guessed the model's URL and exploited shared contractor API keys. Former national cyber director Kemba Walden warned Mythos "can hack nearly anything and we aren't ready," amplifying the debate over concentrated risk in closed-source frontier AI.
Amazon agreed to invest up to an additional $25B in Anthropic (initial $5B tranche), on top of $8B previously committed. Anthropic has pledged to spend over $100B on AWS over the next decade. The deal comes as Anthropic's revenue tripled to $30B and the company approaches a potential IPO that could value it near $800B.
CEO Sundar Pichai announced at Cloud Next 2026 that AI generates 75% of Google's new code, up from 25% in October 2024. Engineers have shifted to a reviewer role, and agent-assisted workflows completed a complex code migration six times faster than a year ago. This is the most concrete data point yet on AI's displacement of traditional software engineering workflows at scale.
Meta announced layoffs of roughly 10% of staff (about 8,000 people) effective May 20, alongside closing 6,000 open roles — all to fund a 2026 AI capex of $115–135B, nearly double last year's $72B. The company remains capacity-constrained with compute demand outpacing supply. The layoffs underscore big tech's brutal calculus: fewer humans, more GPUs.
Anthropic released a detailed postmortem on April 23 confirming three separate bugs degraded Claude Code from March through April: reasoning effort silently dropped, a caching bug wiped thinking history, and a system prompt change capped responses to 100 words. All three are fixed in v2.1.116, and Anthropic reset usage limits for all subscribers — a notable act of transparency in an industry that typically stays quiet about quality regressions.
Google debuted its Gemini Enterprise Agent Platform on April 22, unifying model selection, agent building, orchestration, DevOps, and security under one roof. It provides access to 200+ models including Gemini 3.1 Pro and Gemma 4, and is generally available immediately. Paired with a $750M agentic AI partner fund targeting 120,000+ partners, Google is making its clearest enterprise play yet.
DeepSeek is in talks with Tencent and Alibaba to raise at least $300M in its first external funding round. Tencent proposed a 20% stake; DeepSeek pushed back on ceding that much control. The target valuation doubled from $10B to $20B+ in just 48 hours — a testament to the frenzy around the Chinese lab that trained frontier models at a fraction of Western costs.
Mira Murati's Thinking Machines Lab signed a deal worth single-digit billions with Google Cloud for AI infrastructure powered by Nvidia's latest GB300 chips. The startup will be among the first to use Google's A4X Max VMs. Their product "Tinker" automates creation of custom frontier models — positioning the former OpenAI CTO's startup as a serious infrastructure-scale player barely a year after founding.
