Zoom announced that its “federated” AI system scored 48.1% on Humanity’s Last Exam, an expert-level reasoning benchmark. Zoom says the approach orchestrates multiple top models (including from OpenAI, Anthropic, and Google) alongside Zoom’s own smaller models using a “Z-scorer” selection method. According to the post, the 48.1% result edges out the previous leader Gemini 3 Pro’s 45.8% — though Zoom notes the leaderboard has continued to move, with GPT-5.2 later reaching 50%. Zoom says this capability will power AI Companion 3.0, aiming to deliver more accurate summaries, stronger reasoning, and broader task automation across the platform. Zoom also notes that the claim has faced scrutiny from another AI tool (DeepWriter), which reported a higher score (50.91%) in late November. Overall, the announcement highlights an emerging “federated” template where enterprises combine multiple frontier models to deliver cutting-edge AI outcomes without training a single system end-to-end.