Claude Runs 1,000 Agents at Once; Google Ships One Gemini Agent
Anthropic's Claude runs up to 1,000 sub-agents at once, Google Cloud ships one enterprise Gemini agent, and NVIDIA trains agents to recover from mistakes.
Agents took most of the day. Anthropic and Google both shipped agent products, and NVIDIA published a training method aimed at multi-turn agents.
-
Anthropic Claude dynamic workflows. Claude Managed Agents can now split a job across up to 1,000 sub-agents running in parallel, with a lead agent handing out the tasks. In Anthropic’s testing, one agent found at most 27 of 70 hidden bugs in a codebase, while the multi-agent workflow caught 66. The results come from Anthropic’s own test, and the report gives no price. Read more
-
Google Cloud Gemini agent. A cloud-hosted agent for enterprise work that answers questions, does knowledge work, creates media, and writes and runs code. It runs from one prompt box and one API. The report does not say what it costs. Read more
-
Anthropic OSS Scanner. A free service in which open-source projects that opt in get periodic security scans from Anthropic’s strongest models. The report hints at a trade-off but the text cuts off before naming it. Read more
-
NVIDIA PivotOPD. An on-policy distillation method, meaning the model learns from its own live outputs, that trains multi-turn LLM agents to avoid early pivotal mistakes and recover from them. It posted the best average against 13 baselines on 3 agent benchmarks. The results cover only those three benchmarks so far. Read more
