Google Gemini 1.0
Google and DeepMind unveiled Gemini 1.0 as their 'largest and most capable' model, built natively multimodal to reason across text, images, audio, video, and code from the ground up. It came in three sizes (Ultra for complex tasks, Pro for general scaling, Nano for on-device), and Google claimed Gemini Ultra was the first model to beat human experts on the MMLU benchmark and to outperform GPT-4 on most leading tests. It was presented as Google's true frontier entry, not a fast-follow.
Gemini Pro went live in Bard on launch day and became available to developers via Google AI Studio and Vertex AI on December 13, 2023, while Gemini Nano shipped on the Pixel 8 Pro. The flagship Gemini Ultra did not ship at announcement; it arrived in early 2024 inside a paid 'Bard Advanced' tier.
Gemini 1.0's reputation was defined less by the model than by a credibility self-inflicted wound. The viral 'Hands-on with Gemini' video appeared to show the model reacting to live video and voice in real time, but Google later confirmed it was assembled from still image frames and text prompts with the spoken interaction dubbed in and latency edited out. The underlying benchmark wins were also narrow and partly methodology-dependent, and the headline Ultra model was not actually available on announcement day. For an executive audience, this is the canonical 'claims versus reality' launch: a capable model genuinely competitive with GPT-4, undercut by a demo that implied capabilities the product did not yet deliver.
It shows how a staged demo can overshadow real technical progress, and why leaders should treat polished launch videos as marketing artifacts to be verified, not as evidence of shipped capability.