Xuedong Huang

CTO@Zoom

[Xuedong Huang @xdh]

https://t.co/JKXi4qAzC2

[Xuedong Huang @xdh]

Zoom AI Scribe API open to all developers https://t.co/MKIkatsDvh

[Xuedong Huang @xdh]

Zoom’s amazing customer support agentic AI breakthrough https://t.co/BssIZVsDio

[Xuedong Huang @xdh]

https://t.co/n0v96p43iL

[Xuedong Huang @xdh]

Zoom’s federated AI engine can scale up and scale out on the hardest SOTA https://t.co/edpj5BATyW

[Xuedong Huang @xdh]

https://t.co/TpVAfDIbhk

[Xuedong Huang @xdh]

Check out https://t.co/NhXMFiiPii https://t.co/iSn35meLmG

[Xuedong Huang @xdh]

There’s been increasing discussion around Zoom AI’s leading performance on Humanity’s Last Exam (HLE)—one of the most demanding benchmarks in AI today. I want to step back and highlight what truly matters here: this is not just a strong technical result, but a shift in how AI should be built and deployed in the real world.

For too long, progress in AI has been framed as a competition between individual models—their size, parameters, and raw capabilities. But in enterprise reality, value doesn’t come from having the “best” model. It comes from how well models are orchestrated: the ability to reason across steps, validate outputs, correct errors, and reliably drive work to completion in complex environments.

That’s why Zoom never bet on a single model. From the beginning, we’ve focused on building a federated, multi-agent AI system—one where multiple frontier models collaborate, cross-check one another, and complement each other’s strengths. This system-level approach is what enables Zoom AI to outperform any standalone model on a growing set of challenging tasks.

Humanity’s Last Exam is not a one-off result. The same methodology has delivered strong outcomes across other difficult evaluations as well, reinforcing our belief that the future of AI is collaborative, resilient, and built for real work—not just benchmarks.

[Xuedong Huang @xdh]

@BanghuaZ Watch our new AI Companion 3.0 GA

[Xuedong Huang @xdh]

Zoom AI teams achieved a new state-of-the-art score of 48.1% on the highly challenging Humanity’s Last Exam (HLE), marking a major breakthrough for our Federated AI strategy. We have unlocked stronger capabilities in exploration, reasoning, and multi-model collaboration, surpassing the performance limits of any single model. https://t.co/X6du2MU37S

[Xuedong Huang @xdh]

Zoom closed captions being enabled to all Zoom customers globally. Another AI for accessibility milestone!

https://t.co/fXGai3MM3F

[Xuedong Huang @xdh]

https://t.co/loSZCt3AVJ

[Xuedong Huang @xdh]

Speech recognition is now across the full loop: client to cloud and back. https://t.co/mFhZpUpP3E

[Katherine Fung @katherinekfung]

Hear what Huang has to say when asked what strategies Zoom has employed to encourage adoption of its AI tools 👇

[Xuedong Huang @xdh]

https://t.co/ePEZGpOeJA

[Xuedong Huang @xdh]

@Zoom These are the SLM benchmarks in the 2b category

[Zoom @Zoom]

Zoom CTO @xdh gives his thoughts on how small language models (SLMs) drive our vision for AI agents that work together in a federated approach to improve your day-to-day tasks 💡➡️ https://t.co/n1t0f0cTWB

As artificial intelligence continues to improve, we are preparing for the next iteration: agentic AI ✨

Read the blog to learn how Zoom’s new SLMs stack up against today’s LLMs, how we’re setting Zoom up for the agentic AI era, and what it means for you ✅

[Xuedong Huang @xdh]

Microsoft Azure AI used to be the unmatched leader in speech recognition as published by Standford AI Index in 2021. It is rewarding to see Zoom outperformed other leaders in 2024 in our pushing to help our customers with AI Companion 2.0. https://t.co/CTNcHcysSs

[Xuedong Huang @xdh]

https://t.co/7YU8RKewRi