Google announced Gemini 1.0 on December 6, 2023, as a family of three multimodal models: Ultra for the most complex tasks, Pro for a wide range of tasks, and Nano for on-device use. Only Pro and Nano shipped that day, in the Bard chatbot and on the Pixel 8 Pro phone, while Ultra waited for further safety checks. The launch came with a headline benchmark claim and a demo video that both drew scrutiny, and the Gemini line has been replaced several times since, most recently by Gemini 4 Argon in October 2026.
Key takeaways
- Gemini 1.0 was announced on December 6, 2023, with Ultra, Pro, and Nano sizes.
- Google reported 90.0 percent for Ultra on the MMLU benchmark, a score that depended on a special prompting method.
- A launch demo video was edited, and Google acknowledged it did not show live voice interaction.
- Ultra reached consumers on February 8, 2024, when Bard was renamed Gemini.
- Gemini 3 Pro arrived on November 18, 2025, Gemini 3.5 Flash in May 2026, and Gemini 4 Argon became generally available in the Gemini API on October 13, 2026.
What Google announced on December 6, 2023
The announcement on Google's blog carried notes from chief executive Sundar Pichai and Demis Hassabis of Google DeepMind. Google described Gemini as built to be multimodal from the start, meaning one model that handles text, code, audio, images, and video. Ultra was the largest size, Pro was positioned for scaling across many tasks, and Nano was the most efficient version for phones and similar devices.
Availability was staggered. A fine-tuned Gemini Pro began running in Bard that day, in English in more than 170 countries and territories. Gemini Nano powered features such as Summarize in the Recorder app and Smart Reply in Gboard on the Pixel 8 Pro. On December 13, 2023, developers and enterprises could reach Gemini Pro through the Gemini API in Google AI Studio or Vertex AI. Ultra went first to select customers and testers, with a wider release promised for early 2024.
The benchmark claim and its fine print
Google said Ultra exceeded prior best results on 30 of 32 widely used academic benchmarks and was the first model to outperform human experts on MMLU, a test spanning 57 subjects. The company's technical report, posted to arXiv on December 19, 2023, gives Ultra 90.04 percent, against a human-expert level of 89.8 percent as gauged by the benchmark authors.
The report also shows how that number was reached. It used a chain-of-thought method with 32 samples that accepts the majority answer only when agreement passes a threshold and otherwise falls back to a single direct answer. Under the more common five-shot setup, the same table lists Ultra at 83.7 percent and GPT-4 at 86.4 percent, while under the 32-sample method GPT-4 is listed at 87.29 percent. Both comparisons are Google's own reported figures, and they use different prompting conditions, so a headline gap between models depends on which row is read.
The demo video
A launch video showed a person interacting with Gemini by voice and gestures, and it spread widely. An opinion columnist, Parmy Olson reported on December 7, 2023, that Google had acknowledged the video was edited and that Gemini had been prompted with still frames and written text rather than live spoken input. Engadget summarized that account on December 8. The video description said latency had been reduced and outputs shortened, but it did not state the input method. Oriol Vinyals, a Gemini co-lead at DeepMind, said on X that the prompts and outputs shown were real and that the video was meant to show what multimodal experiences could look like.
Timeline of what followed
The table covers releases that can be traced to Google's own posts. It is selective and ends where verified sources end.
| Date | Event | Detail from Google |
|---|---|---|
| December 6, 2023 | Gemini 1.0 announced | Pro in Bard, Nano on Pixel 8 Pro, Ultra pending |
| February 8, 2024 | Bard renamed Gemini; Ultra 1.0 offered in Gemini Advanced | Included in the Google One AI Premium plan at USD 19.99 per month, with a two-month free trial |
| February 15, 2024 | Gemini 1.5 Pro introduced | Limited preview; 128,000-token standard window, up to 1 million tokens experimentally |
| November 18, 2025 | Gemini 3 Pro launched in preview | Available in the Gemini app, AI Mode in Search, the Gemini API, and Vertex AI |
| May 2026 | Gemini 3.5 Flash made generally available at Google's developer conference | Default model for AI Mode globally, per Google's roundup dated May 20, 2026 |
| July 21, 2026 | Gemini 3.6 Flash generally available | Released with Gemini 3.5 Flash-Lite, per the Gemini API changelog |
| September 2, 2026 | Gemini 3.8 Flash generally available | Listed in the Gemini API changelog |
| October 13, 2026 | Gemini 4 Argon generally available | Listed in the Gemini API changelog; the models page shows it as Stable |
In that May 2026 post, Google said Gemini 3.5 Pro was already in internal use and expected the following month. As of October 27, 2026, the Gemini API changelog and models page list no Gemini 3.5 Pro model, and the changelog says the 3.5 Flash release was deprecated on October 8, 2026, with requests routed to Gemini 3.6 Flash. Gemini 3 Pro Preview is listed as shut down. Benchmark scores quoted by Google for newer models are self-reported and are not repeated.
How to read launch announcements
Three habits help with any model launch. First, separate what is available today from what is promised: Ultra was announced in December 2023 and reached the public two months later. Second, look for the evaluation method behind a score, since prompting choices change results. Third, treat a demo as an illustration of intended use. Google's technical report and blog post are the primary documents for Gemini 1.0, and they remain the place to check claims.
FAQ
When was Google Gemini announced?
December 6, 2023. Gemini Pro came to Bard that day, and the API opened to developers on December 13, 2023.
What were the three Gemini sizes?
Ultra, Pro, and Nano. Ultra was the largest, Pro a general-purpose middle size, and Nano a small model for on-device tasks.
Did Gemini Ultra beat GPT-4?
Google reported higher scores on many benchmarks. The MMLU headline relied on a 32-sample prompting method, and in the five-shot comparison listed in Google's own report, GPT-4 scored higher than Ultra.
Is Bard still a product?
No. Google renamed Bard to Gemini on February 8, 2024.
Related reading
- Character AI and Google Cloud
- Google's earlier LaMDA language model
- The GPT-3 language model
- A deep learning model with over one trillion parameters
More in AI news.
Last verified: October 2026
Sources
- Pichai S and Hassabis D, Introducing Gemini: Google’s most capable AI model yet, Google, December 6, 2023
- Gemini Team, Gemini: A Family of Highly Capable Multimodal Models, arXiv, December 19, 2023
- Google Bard is now Gemini: How to try Ultra 1.0 and new mobile app, Google, February 8, 2024
- Introducing Gemini 1.5, Google’s next-generation AI model, Google, February 15, 2024
- Gemini 3: Introducing the latest Gemini AI model from Google, Google, November 18, 2025
- Google roundup of its May 2026 developer conference announcements, May 20, 2026
- Gemini API release notes (changelog) and models page, Google AI for Developers, checked October 27, 2026
- Engadget, Google admits that a Gemini AI demo video was staged, December 8, 2023