AI DAILY / DEV
Weekly Rollup
Week 16

Claude Mythos: 93.9% SWE-bench, Found Zero-Days in Every Major OS

Anthropic confirmed Claude Mythos, scoring 93.9% on SWE-bench Verified and 97.6% on USAMO. The model autonomously discovered zero-day vulnerabilities in every major OS and browser during testing, including a 27-year-old OpenBSD bug. Access restricted to 50 organizations under Project Glasswing.

models infoq.com

OpenAI Splits Codex into Basic and Advanced Tiers

OpenAI restructured Codex with separate Basic and Advanced tiers, adding app preview, scratchpad, a Pull Requests section, and inline comments on previewed UI. The split gives power users deeper integration while keeping the entry point accessible.

tools openai.com

Microsoft's markitdown Converts Any Doc to LLM-Ready Markdown

Microsoft open-sourced markitdown, converting PDFs, DOCX, PPTX, HTML and other formats into clean Markdown suitable for LLM context windows. Gained 3,600+ stars in its first week. Solves a pain point every RAG pipeline developer has hit.

open-source github.com

Google ADK Becomes the Most Complete Multi-Agent Framework

Google's Agent Development Kit (adk-python) hit 8,200+ stars and emerged as the most fully-featured multi-agent framework this month. Meanwhile Block's Goose takes a different approach: local-first with native MCP support for tool integration.

frameworks github.com

Gemma 4 Goes Viral — Developers Test Google's Best Open Model on Claude Code

Gemma 4's Apache 2.0 release sparked a wave of testing across the developer community. Medium posts and Reddit threads show it scoring competitively on code tasks when used with Claude Code. The 31B model's 256K context and zero licensing cost are driving rapid adoption.

models blog.google

Claude Code Epitaxy — Anthropic's Desktop Overhaul Leaked

Anthropic is building a major Claude Code desktop overhaul codenamed Epitaxy, featuring Coordinator Mode for parallel sub-agents, plan/task/diff panels, multi-repo support, and code preview. The power-user UI suggests Claude Code is evolving from a CLI into a full IDE.

tools geeky-gadgets.com

Anthropic Valued at $800B — Revenue Surpasses OpenAI

Anthropic has received VC offers at valuations up to $800B, more than doubling its $380B February valuation. With annualized revenue at $30B vs OpenAI's $25B, it's the first time Anthropic leads on revenue. Over 1,000 enterprises spend $1M+/year on Claude.

industry cnbc.com

Claude Code Ecosystem Explodes on GitHub — 4 Repos Trending Simultaneously

Four Claude Code extension repos are trending on GitHub this week, collectively gaining 8K+ stars: andrej-karpathy-skills, claude-code-best-practice, claude-mem, and learn-claude-code. The pattern signals Claude Code is becoming a platform with its own ecosystem, similar to VS Code extensions circa 2018.

community github.com

OpenAI Launches GPT-5.4-Cyber for Defensive Security

OpenAI released GPT-5.4-Cyber, a model fine-tuned for defensive cybersecurity, days after Anthropic's Mythos disclosure. Currently scaling to highest-tier API customers. The cybersecurity AI arms race between Anthropic and OpenAI is heating up fast.

models startupnews.fyi

Google Open-Sources Magika: File Detection Trained on 100M Files

Google open-sourced Magika, its internal AI file type detection tool with 99% accuracy across 200+ content types. Previously used to scan hundreds of billions of files weekly across Gmail and Drive. Now available for developers to build into their own pipelines.

open-source github.com