Claude Opus 4.6 → 4.7: agent teams in Claude Code + 1M context + Cyber Verification Program

[claude-yolo]

WHY users should care

Two new flagship models dropped in quick succession — both have Claude Code implications.

🚀 Claude Opus 4.6

• Agent teams in Claude Code — assemble teams of agents on tasks (new Claude Code feature)
• 1M token context window (beta)
• Compaction — model summarizes its own context for longer runs
• Adaptive thinking — auto-picks when to use extended thinking
• Effort controls — /effort param (high/medium/low) to tune intelligence vs. cost
• Model ID: claude-opus-4-6

🚀 Claude Opus 4.7

• Better coding on hard long-horizon tasks (+13% on 93-task benchmark per Cursor)
• Substantially better vision (higher resolution image support)
• Cyber Verification Program — security researchers can apply for enhanced cyber capabilities; auto-safeguards on by default otherwise
• Model ID: claude-opus-4-7

Also in this batch

• 3 new research papers: trustworthy-agents, values-wild, transformer-circuits
• Settings docs: minor /status UI clarifications (non-breaking)

Source PR

thevibeworks/claude-code-docs#1044

@lroolle — two releases worth knowing about, especially the Claude Code agent teams feature.

💤 auto-filed by claude-yolo day shift