Google today introduced Gemini 1.0, its most capable AI model yet, claiming state-of-the-art performance on 30 of 32 widely-used academic benchmarks. The model comes in three sizes: Ultra for highly complex tasks, Pro for scaling across a wide range of tasks, and Nano for on-device. Gemini Ultra scored 90.0% on the MMLU benchmark, the first model to outperform human experts. The model is described as ‘natively multimodal,’ trained from the start on text, code, audio, image and video.

Gemini Pro is already powering Bard, Google’s chatbot, in what the company calls the biggest upgrade since Bard launched. The Pro version will be available to developers via API on December 13. Gemini Nano runs on the Pixel 8 Pro, enabling features like Summarize in Recorder and Smart Reply in Gboard. Gemini Ultra is undergoing safety checks and will be available to select customers early next year, followed by a new ‘Bard Advanced’ experience.

Some observers remain skeptical. Analyst Andy Thurai noted Google is ‘clearly playing catch-up with OpenAI’ and pointed to the delayed broader rollout of Ultra as a sign of caution. The company emphasized responsibility, calling Gemini’s safety evaluations the most comprehensive for any Google model to date.