Why Salesforce's CEO has switched from ChatGPT to Gemini 3
What's the story
Salesforce CEO Marc Benioff has announced his switch from OpenAI's ChatGPT to Google's latest artificial intelligence (AI) model, Gemini 3. The tech mogul, who has been using ChatGPT daily for the past three years, was blown away by the performance of Gemini 3 after spending two hours testing it. He took to X to share his thoughts on the upgrade.
User experience
Benioff's reaction to Gemini 3
Benioff was clearly impressed by Gemini 3, calling the upgrade "insane" in terms of reasoning, speed, and multimodal capabilities. He wrote on X, "Holy s**t. I've used ChatGPT every day for three years. Just spent two hours on Gemini 3. I'm not going back." He further added that the leap is insane across all aspects including reasoning, speed, images, and video with everything feeling sharper and faster.
Model performance
Gemini 3's capabilities and industry response
Gemini 3 was unveiled last week by Google. The model can generate and understand text, images, video, and code with deep integration across the Google ecosystem. Google CEO Sundar Pichai called it the company's "most intelligent model," while OpenAI CEO Sam Altman congratulated Google on the launch. Former Tesla AI director Andrej Karpathy also praised Gemini 3 as a "tier 1 LLM" with strong potential for daily use.
Benchmark results
Gemini 3's performance surpasses competitors
According to a Wall Street Journal report, Gemini 3 has surpassed OpenAI's ChatGPT and other leading models across standard industry benchmarks. The model was tested privately by Google employees for months before its public debut. Tulsee Doshi, Senior Director of Product Management for Gemini, revealed that one of her informal tests showed significant improvement in the system's ability to write in Gujarati, an Indian language with less online prominence.
Evaluation feedback
Early evaluations of Gemini 3 yield strong results
Box CEO Aaron Levie revealed his company's early evaluation of Gemini 3 yielded surprisingly strong results. After testing large and complex document sets, the cloud content management firm found the model scoring significantly higher than expected. The performance jump was so big that his team doubted at first whether they had run their assessments correctly.