Story cards stay below the sticky dock while audio, chapters, date, and brief navigation remain accessible.
Anthropic is internally testing a frontier model codenamed "Claude Mythos," described by insiders as a "step change in capabilities" over current Claude models. A release is expected sometime in April or Q2 2026. Separately, Anthropic's Model Context Protocol hit 97 million installs in March — signaling MCP has crossed from experiment to foundational agent infrastructure.
Source: [llm-stats.com](https://llm-stats.com/ai-news)
---
Elon Musk announced a merger between SpaceX and xAI, embedding Grok models directly into SpaceX operations to drive autonomous spacecraft and Mars colony robotics. The deal blurs the line between AI lab and aerospace company in a way no one has attempted before.
Source: [crescendo.ai](https://www.crescendo.ai/news/latest-ai-news-and-updates)
---
Google's Gemini 3.1 Pro scored 94.3% on GPQA Diamond, topping reasoning benchmarks. Google also shipped Gemini 3.1 Flash-Lite — 2.5x faster, 45% quicker output, priced at $0.25 per million input tokens — a direct shot at cost-sensitive enterprise buyers.
Source: [llm-stats.com](https://llm-stats.com/llm-updates)
---
OpenAI surpassed $25 billion in annualized revenue and is taking early steps toward a public listing as soon as late 2026. GPT-5.4 (released March 5) is driving the growth, posting record scores on computer-use and knowledge-work benchmarks.
Source: [llm-stats.com](https://llm-stats.com/ai-news)
---
Oracle announced plans to raise up to $50 billion to fund a massive expansion of AI data center infrastructure. It's the largest single infrastructure commitment from an enterprise software company yet, underscoring how the compute arms race shows no signs of cooling.
Source: [solutionsreview.com](https://solutionsreview.com/artificial-intelligence-news-for-the-week-of-april-2-updates-from-datacamp-insight-jam-learnupon-more/)
---
Ledger CTO Charles Guillemet warned that AI is compressing the skill and time required to execute crypto exploits — tasks that once took months of expert research now take minutes with the right prompts. This comes on the heels of $1.4 billion in crypto losses from hacks over the past year.
Source: [CoinDesk](https://www.coindesk.com/tech/2026/04/05/ai-is-making-crypto-s-security-problem-even-worse-ledger-cto-warns)
---
Meta's next-gen model pair — Mango (image/video generation) and Avocado (LLM for text and coding) — are being positioned as a major inflection point for Meta AI. Both are expected in H1 2026, putting pressure on OpenAI and Google in consumer and developer markets.
Source: [MIT Sloan ME](https://www.mitsloanme.com/article/meta-readies-new-ai-models-to-take-on-openai-and-google/)
---
**Big Picture**
Today's news is a study in scale and consolidation. The frontier model race is hitting a new gear simultaneously across every major lab — Anthropic with Mythos, Google with Gemini 3.1, Meta with Mango/Avocado — while the money side of the industry is entering a new phase: $25B ARR at OpenAI, $50B infrastructure raises at Oracle, a SpaceX-xAI merger. The era of AI as an experimental bet is over; it's now a full-contact capital war. The wild card is the security warning from Ledger — as these models get more powerful, the attack surface for everyone grows with them.
