🤖 AI Agents Weekly: Claude Fable 5, Kimi K2.7-Code, NotebookLM Goes Agentic, DiffusionGemma, MiMo Code, and More
Claude Fable 5, Kimi K2.7-Code, NotebookLM Goes Agentic, DiffusionGemma, MiMo Code, and More
In today’s issue:
Anthropic ships Mythos-class Claude Fable 5
Kimi K2.7-Code open-sources a 1T coder
NotebookLM becomes an agentic workstation
Google’s DiffusionGemma generates text in blocks
Xiaomi open-sources MiMo Code agent
Cohere ships North Mini Code
Gemini 3.5 Live Translate goes real-time
Gemini-SQL2 tops BIRD text-to-SQL
Apple rebuilds Siri on Google Gemini
Grok opens a plugin marketplace
Claude Code adds nested subagents
Nex-N2 opens an agentic model series
Extend UI ships document-agent components
Cognition’s FrontierCode raises the eval bar
Study questions the multi-agent advantage
Recursive automates AI research
And all the top AI dev news, papers, and tools.
Top Stories
Anthropic Launches Claude Fable 5, Its First Public Mythos-Class Model
Anthropic released Claude Fable 5, a Mythos-class model made safe for general use, alongside the restricted Claude Mythos 5. It is the most capable model Anthropic has ever made widely available, and its lead grows the longer and more complex the task.
State-of-the-art across the board: Fable 5 tops nearly every tested benchmark, with the widest margins on long, multi-step reasoning and autonomous task completion.
Software engineering: Posts SOTA on Cognition’s FrontierCode, and Stripe reported it compressed a 50-million-line codebase migration from two months of human work into a single day.
Agentic and vision: Holds focus across millions of tokens, runs roughly 3x better on strategic gameplay with persistent memory, and finished Pokémon FireRed from raw screenshots with no helper tools.
Safeguards by fallback: Requests touching cybersecurity, biology, chemistry, or distillation are routed to Claude Opus 4.8 instead of refused, triggering in under 5% of sessions. Mythos 5 stays restricted to Project Glasswing partners.
Pricing: $10 per million input tokens and $50 per million output tokens, with rollout across plans continuing through June 22.