Zoom's AI Tops Humanity's Last Exam, But Critics Question the Method Behind the Score

The San Jose-based company Zoom Video Communications announced last week that it had achieved the highest score ever recorded on one of artificial intelligence's most demanding tests, the Humanity's Last Exam. According to Zoom, its AI system scored 48.1 percent on the benchmark, edging out Google's Gemini 3 Pro, which held the previous record at 45.8 percent. The company's chief technology officer, Xuedong Huang, wrote in a blog post that Zoom has achieved a new state-of-the-art result on the challenging Humanity's Last Exam full-set benchmark, scoring 48.1, which represents a substantial 2.3 improvement over the previous state-of-the-art result. Zoom's achievement has sparked both surprise and skepticism within the technology industry, with many questioning how a video conferencing company with no public history of training large language models was able to vault past Google and OpenAI. "It's a remarkable achievement, but it's also a bit puzzling," said Dr. Andrew Ng, a well-known AI expert and former head of AI at Google. "Zoom has been focused on video conferencing and not on developing large language models, so it's unclear how they were able to achieve this result." The Humanity's Last Exam is a benchmark designed by subject-matter experts worldwide to stump even the most advanced AI models. The test assesses an AI system's ability to reason, understand, and generate human-like language, and is considered one of the most challenging AI benchmarks available. According to Dr. Ng, the fact that Zoom was able to achieve a higher score than Google's Gemini 3 Pro suggests that the company may have been using a different approach to training its AI system. Zoom has not disclosed the specifics of its AI training methodology, leading some to speculate that the company may have been using a combination of pre-trained models and fine-tuning techniques to achieve its results. "It's possible that Zoom is using a combination of pre-trained models and fine-tuning techniques to achieve its results," said Dr. Ng. "But without more information, it's difficult to say for sure." The implications of Zoom's achievement are significant, as it suggests that even companies with limited experience in AI development can make significant breakthroughs in the field. However, the achievement also raises questions about the potential for AI systems to be trained on proprietary data and methods, which could potentially give companies an unfair advantage in the market. As the AI industry continues to evolve, it remains to be seen how Zoom's achievement will impact the development of AI systems in the future. The company has not announced any plans to release its AI system or make it available for public use, leading some to speculate that the achievement may be more of a publicity stunt than a genuine breakthrough. However, the achievement has undoubtedly sparked a new wave of interest in AI development, and it will be interesting to see how the industry responds to Zoom's announcement in the coming weeks and months.