Jump to a brief
Grouped by topic first; AM/PM are editions inside the topic.
AI News Afternoon Briefing — 2026-04-04 10:28
---
**GPT-4o Is Dead; GPT-5.4 Takes the Throne**
OpenAI quietly retired GPT-4o on April 3 and is now rolling out GPT-5.4 Thinking to Plus, Team, and Pro users. The new model is the first general-purpose reasoning model with native computer-use at up to 1M context tokens, and it scores 75% on OSWorld-Verified — a 28-point jump over GPT-5.2.
Source: OpenAI / TechCrunch — https://techcrunch.com/2026/03/05/openai-launches-gpt-5-4-with-pro-and-thinking-versions/
---
**OpenAI Closes a $122B Round at an $852B Valuation**
OpenAI announced a massive funding round, including $3B from retail investors, cementing its position as the most valuable private company on earth. The capital is earmarked for frontier compute, global expansion, and enterprise ChatGPT and Codex demand.
Source: OpenAI — https://openai.com/index/accelerating-the-next-phase-ai/
---
**Anthropic's Mythos 5 Is a 10-Trillion-Parameter Cybersecurity Wildcard**
Anthropic launched Claude Mythos 5, a 10T-parameter model that experts say can find and exploit software vulnerabilities at an unprecedented scale. CNN reports that security researchers are treating the release as both a defensive asset and a serious threat vector.
Source: CNN Business — https://edition.cnn.com/2026/04/03/tech/anthropic-mythos-ai-cybersecurity
---
**Anthropic Accidentally Published 500K Lines of Claude Code to npm**
A misconfigured debug file caused Anthropic to ship nearly its entire Claude Code source tree — ~2,000 files, ~500K lines — to the public npm registry on March 31. The leak has since been scrubbed, but the window of exposure is drawing scrutiny.
Source: Agentic AI Hype / Substack — https://agenticaihype.substack.com/p/anthropics-500k-line-code-leak-google
---
**DeepSeek V4: Frontier Performance for $5.2M**
DeepSeek released V4, a 1-trillion-parameter Mixture-of-Experts model with fully open weights under Apache 2.0. It benchmarks competitively with top US frontier models at a training cost that is orders of magnitude lower — renewing pressure on labs burning billions per run.
Source: devFlokers — https://www.devflokers.com/blog/ai-news-last-24-hours-april-2026-model-releases-breakthroughs
---
**Google Ships Gemma 4 Under Apache 2.0**
Google DeepMind released Gemma 4, its most capable open model family yet, purpose-built for advanced reasoning and agentic workflows. The Apache 2.0 license makes it fully commercial-use friendly, a strategic move to seed the open ecosystem against Meta's Llama and DeepSeek.
Source: Agentic AI Hype — https://agenticaihype.substack.com/p/anthropics-500k-line-code-leak-google
---
**SpaceX Absorbs xAI in a $250B Deal**
Elon Musk's SpaceX has acquired xAI — the company behind Grok — for $250 billion, folding AI capabilities directly into SpaceX's infrastructure ambitions. The move raises obvious questions about conflicts of interest given Musk's simultaneous roles at DOGE and Tesla.
Source: devFlokers — https://www.devflokers.com/blog/ai-news-last-24-hours-april-2026-model-releases-breakthroughs
---
**Big Picture**
Today's news is a stress test in two directions at once: capability and accountability. The frontier is accelerating faster than the safety infrastructure around it — a 10T-parameter model that can autonomously exploit vulnerabilities drops the same week Anthropic accidentally exposes its own source code. Meanwhile, DeepSeek V4's Apache-licensed, $5M training run is a continuing reminder that the moat every US lab is counting on — raw compute spend — is leaking. The capital is enormous ($122B to OpenAI alone), but so is the fragility.
---
Sources:
- [OpenAI launches GPT-5.4 with Pro and Thinking versions | TechCrunch](https://techcrunch.com/2026/03/05/openai-launches-gpt-5-4-with-pro-and-thinking-versions/)
- [OpenAI raises $122 billion | OpenAI](https://openai.com/index/accelerating-the-next-phase-ai/)
- [Anthropic's next model could be a 'watershed moment' for cybersecurity | CNN Business](https://edition.cnn.com/2026/04/03/tech/anthropic-mythos-ai-cybersecurity)
- [Anthropic's 500K-Line Code Leak, Google Gemma 4 Goes Apache | Agentic AI Hype](https://agenticaihype.substack.com/p/anthropics-500k-line-code-leak-google)
- [AI News Last 24 Hours: April 2026 | devFlokers](https://www.devflokers.com/blog/ai-news-last-24-hours-april-2026-model-releases-breakthroughs)
- [Introducing GPT-5.4 | OpenAI](https://openai.com/index/introducing-gpt-5-4/)