Zoom has topped OpenAI and Google on one of the industry’s toughest AI benchmarks, highlighting the success of its federated approach which combines its own small language models with advanced open-source and closed-source models from other providers for optimal performance.
The company’s AI system, which routes queries to specialized models from OpenAI, Anthropic, Meta, and others based on task requirements, scored 53.0 on Humanity’s Last Exam in December, beating OpenAI’s GPT-5.2 Pro and Google’s Gemini Deep Research.
According to Xuedong Huang, Zoom’s chief technology officer:
“Our success reinforces a fundamental belief: the future of AI is collaborative, not competitive. By combining the best innovations from across the industry with our own research breakthroughs, we create solutions that are greater than the sum of their parts.”
“We’re grateful to Anthropic, Google, and OpenAI for their continued innovation, which makes our breakthroughs possible. Their groundbreaking work provides the foundation upon which we build specialized, efficient solutions tailored to real-world workplace challenges.”
What is Zoom’s Federated AI Architecture?
Launched in October 2025, Zoom’s AI Companion 3.0 combines results from multiple AI models through what the company calls a “Z-scorer” system that selects or refines outputs from multiple providers.
Organizations choose from three deployment models based on compliance requirements. The federated approach uses external AI providers with enterprise security protocols. Two Zoom-hosted options keep processing on company infrastructure with different model access levels, addressing data residency requirements for organizations in regulated industries.
The technical approach behind the HLE achievement uses what Huang describes as an “explore-verify-federate” methodology. The system generates multiple reasoning paths for complex problems, then uses interaction between different AI models to evaluate and refine solutions before selecting an answer.
This produced the 53.0 score on the full HLE benchmark, with a 55.2 score on the text-only subset.
Contact Center Applications Drive Commercial Focus
Zoom’s product development priorities show where the company sees near-term returns. AI Companion integrates into Zoom Contact Center with real-time agent assistance, automated quality management, and intelligent routing. The system handles call summarization, action item extraction, and automated quality scoring.
As Smita Hashim, Zoom’s chief product officer, said at the October launch:
“AI Companion 3.0 will provide deeper insights from their conversations to help them get more done at work and drive better business outcomes, regardless of whether they’re working in Zoom Workplace, in person, or across compatible apps and integrations.”
Contact center interactions span technical troubleshooting, product questions, billing inquiries, and complaint resolution: tasks that draw on different specialized models in Zoom’s federated architecture. Beyond contact centers, the system provides meeting summaries, cross-platform information retrieval, and workflow automation for multi-step business processes. A deep research feature synthesizes information from multiple sources into structured documents.
Zoom’s Enterprise Integration Strategy
AI Companion 3.0 connects to Google Workspace, Microsoft 365, Salesforce, and other enterprise platforms through native integrations. Users access AI capabilities through Zoom Meetings, Team Chat, and connected email clients. The company calls this a “conversational work surface” that operates across multiple applications.
The integration strategy positions the system as a layer across existing tools rather than a replacement, reducing workflow disruption. Enterprise AI depends on data access, which explains Zoom’s focus on connectors to workplace applications where organizations store operational data. The approach competes with Microsoft’s Copilot integration across Office 365 and Google’s Workspace AI features.
What the Results Mean for Enterprise Buyers
For unified communications buyers evaluating AI strategies, the HLE results show Zoom’s multi-model approach can deliver competitive performance on tasks requiring complex reasoning.
The $10 monthly standalone pricing and free tier for existing customers create a lower barrier than Microsoft’s Copilot. However, the federated architecture requires organizations to accept routing queries to external AI providers, or choose Zoom-hosted deployment models that limit access to the full range of specialized models.
