List your company

Gemini: A Family of Highly Capable Multimodal Models

About

This report introduces a new family of multimodal models, Gemini, that exhibit remarkable capabilities across image, audio, video, and text understanding. The Gemini family consists of Ultra, Pro, and Nano sizes, suitable for applications ranging from complex reasoning tasks to on-device memory-constrained use-cases. Evaluation on a broad range of benchmarks shows that our most-capable Gemini Ultra model advances the state of the art in 30 of 32 of these benchmarks - notably being the first model to achieve human-expert performance on the well-studied exam benchmark MMLU, and improving the state of the art in every one of the 20 multimodal benchmarks we examined. We believe that the new capabilities of the Gemini family in cross-modal reasoning and language understanding will enable a wide variety of use cases. We discuss our approach toward post-training and deploying Gemini models responsibly to users through services including Gemini, Gemini Advanced, Google AI Studio, and Cloud Vertex AI.

Key insights

What problem do Gemini models address?

Gemini builds natively multimodal Transformer models that jointly process text, image, audio, and video, aiming for strong generalist capabilities alongside domain-specific reasoning without separate single-modality systems.

How does Gemini differ from prior work?

Gemini Ultra advances state of the art on 30 of 32 benchmarks, including 90.04% on MMLU (first above human-expert 89.8%) and 62.4% on MMMU, beating GPT-4V by over 5 points. Evaluated across text, image, video, and audio benchmarks.

What are the practical uses?

The paper documents deployment through Gemini, Gemini Advanced, Google AI Studio, and Cloud Vertex AI. It describes educational tutoring, coding, summarization, and on-device use cases as benchmarked prototypes rather than exhaustive production deployments.

What limitations do the authors note?

The authors note ongoing hallucination issues, struggles with causal and counterfactual reasoning, benchmark saturation limiting evaluation, and the need for more robust standardized benchmarks. They also state models should not be used downstream without prior safety assessment.

Related benchmarks