Mountain ViewGoogle’s own benchmarks put Claude 9 points ahead of its Gemini 4 Argon.

Argon scored 57.4% to Claude Opus 5.5’s 66.4% on Terminal-Bench 4.0, so Google’s new Gemini Agent can hand hard tasks to Claude.

Announced 8 October at Gemini at Work, the agent works across Gmail, Docs and Chat and picks Gemini or Claude per task.

Google Cloud’s Thomas Kurian says the best model may change every few months, so the agent chooses per task.

Gemini 4 Argon, unveiled a week earlier, is not yet open to the public.

How each outlet framed it
TMTPost
frames Google's Gemini Agent as dependent on Claude for hard tasks, and as a play on Workspace's installed base versus Meta's Muse and OpenAI's Dots

Sources: TMTPost