This web content offers a comprehensive overview of performance percentages for various subjects and challenges using the Gemini 1.5 Pro model. Highlights include Code Natural2Code for Python code generation, Math MATH for challenging math problems, Reasoning GPQA for questions in biology, physics, and chemistry, and Reasoning Big-Bench for multi-step reasoning tasks. Surprisingly, the Audio FLEURS section showcases automatic speech recognition in 55 languages, with lower word error rates being better. Overall, the Gemini 1.5 Pro model consistently performs well across different subject areas and challenges, indicating its versatility and effectiveness in various tasks.
https://deepmind.google/technologies/gemini/flash/