Name: Gemini 1.0 Pro Vision
Author: Google DeepMind

Gemini 1.0 Pro Vision

Google DeepMind · 2023-12-13

Gemini 1.0 Pro Vision (`gemini-pro-vision`) was the dedicated multimodal Gemini API endpoint released on December 13, 2023, accepting text and imagery and returning text. Its published API limits were 12,288 input tokens and 4,096 output tokens (a 16,384-token total sequence); Google's Gemini 1.0 technical report documents SentencePiece tokenization for the family.

Benchmark scores

Benchmark	Score
Capability	113.6
MathVista (mini)	45.2
OSWorld	5.8
OSWorld (self-reported, Screenshot + A11y tree)	3.5
OSWorld (self-reported, Set-of-Mark)	1.1
ScienceQA	79.7
ZeroBench (Main questions, pass@1)	0.0
ZeroBench (Main questions, pass@5)	2.0
ZeroBench (Subquestions, pass@1)	12.4
ZeroBench Main Questions pass^5	0.0