Evening brief

Sunday, June 14

0 people · 0 companies · 5 topics

TL;DR Fable 5 and Mythos 5 are still offline globally under a US export control order with no restore date. Claude Sonnet 4 and Opus 4 API endpoints retire tomorrow at 9AM PT with no auto-failover.

Models & research

  • Fable 5 and Mythos 5 went offline June 12 after a US government export directive, citing a claimed jailbreak where the model reads a codebase to surface software flaws; Anthropic calls it a “misunderstanding,” is complying while pushing back publicly, and has no ETA for restoration – all other Claude models are unaffected. TechCrunch · Anthropic statement · VentureBeat
  • Moonshot AI dropped Kimi K2.7-Code on June 12: open-weight 1T-param MoE (32B active, 384 experts), 256K context window, Modified MIT license; self-reported +21.8% on its Kimi Code Bench v2 and beats Claude Opus 4.8 on MCPMark Verified (81.1% vs 76.4%) with ~30% fewer reasoning tokens; no independent third-party numbers yet. MarkTechPost

Tools & dev

  • Hard deadline tomorrow: claude-sonnet-4-20250514 and claude-opus-4-20250514 retire June 15 at 9AM PT; calls fail with errors, no auto-failover to a newer model – update the model string to claude-sonnet-4-5 or claude-opus-4-7 now. MindStudio guide · migration playbook
  • Also June 15: Claude Code Agent SDK and claude -p usage moves off the interactive credit pool into a separate monthly Agent SDK credit bucket. migration playbook
  • If Fable 5 stays suspended, it drops from Pro/Max/Team/Enterprise subscription plans June 22-23 and will require usage credits after that. Developers Digest

OSS

  • addyosmani/agent-skills hit 1,514 stars today: production-grade engineering workflows for coding agents covering API design, frontend, security review, and more in composable layers. GitHub
  • obra/superpowers at 924 stars today: agentic dev methodology from Prime Radiant that forces spec-then-sign-off before execution, works with Claude Code, Codex, Cursor, and other agents. GitHub
  • NVIDIA/SkillSpector at 804 stars today: open-source security scanner for AI agent skills, checking 64 vulnerability patterns across 16 categories via static analysis plus optional LLM semantic pass. GitHub
  • LMCache/LMCache at 238 stars today: persistent KV cache layer for vLLM and SGLang claiming 6.7x faster time-to-first-token and ~80% higher throughput for long-context and RAG workloads. GitHub