Two flagships shipped in one week: Claude Fable 5.2 from Anthropic (Nov 9) and Gemini 4 Carbon from Google (Nov 10). Both are designed not for chat but for multi-step work: coordinating several agents, long tool chains and task delegation.

Claude Fable 5.2 and Mythos 5.2

Anthropic describes the pair as models for “ambitious agentic work”: improved coding, multi-step reasoning and robotics control. Scenarios where the model coordinates and delegates to other agents are called out explicitly — an orchestrator role, not just an executor.

Gemini 4 Carbon

Google shipped Carbon a day later and wired it into Antigravity right away. The /swarm command appeared: a coordinator distributes work to specialised agents and verifiers in isolated git worktrees with a shared message board — a proper answer to the main problem of parallel work, conflicting changes.

What it means

  • Flagships stopped competing on benchmark IQ and started competing on orchestration — a new axis of competition.
  • For agent projects, peak IQ matters less than holding a long plan and delegating well.
  • Both models are available via API — you can start testing today.