Gemini: A Family of Highly Capable Multimodal Models

Benchmark Model Rank Results
visual-question-answering-vqa-on-ai2dGemini Ultra#1EM: 79.5
long-context-understanding-on-mmneedleGemini Pro 1.0#41 Image, 4*4 Stitching, Exact Accuracy: 24.78…
temporal-casual-qa-on-next-qaGemini Ultra (zero-shot)#4WUPS: 29.9
temporal-casual-qa-on-next-qaGemini Pro (zero-shot)#5WUPS: 28.0
video-question-answering-on-perception-test-testGemini Ultra#5Accuracy: 54.7
multi-task-language-understanding-on-mgsmGemini Ultra (8-shot)#6Average (%): 79.0
visual-question-answering-on-mm-vet-v2Gemini Pro Vision#7GPT-4 score: 57.2±0.2
visual-question-answering-vqa-on-infographicvqaGemini Ultra (pixel only)#7ANLS: 80.3
visual-question-answering-vqa-on-textvqa-valGemini Ultra#12Accuracy: 82.3
sentence-completion-on-hellaswagGemini Ultra (10-shot)#22Accuracy: 87.8
visual-question-answering-on-mm-vetGemini 1.0 Pro Vision (gemini-pro-vision)#28GPT-4 score: 64.3±0.4
visual-question-answering-on-docvqa-testGemini Ultra#37ANLS: 0.909
multimodal-reasoning-on-mmmuGemini Ultra#61Accuracy: 59.4
chart-question-answering-on-chartqaGemini Ultra#671:1 Accuracy: 80.8
code-generation-on-humanevalGemini Ultra#75Pass@1: 74.4
math-word-problem-solving-on-mathGemini Ultra (4-shot)#75Accuracy: 53.2
math-word-problem-solving-on-mathGemini Pro (4-shot)#133Accuracy: 32.6
multimodal-reasoning-on-mathvistaGemini Ultra#105Accuracy: 53.0
arithmetic-reasoning-on-gsm8kGemini Pro (maj1@32)#109Accuracy: 86.5