ArchitectureEvaluation
Gemini 1.5: Unlocking Multimodal Understanding Across Millions of Tokens of Context
Google·March 8, 2024
Gemini Team
View on arXivTL;DR
Describes Gemini 1.5 Pro, a mixture-of-experts multimodal model that maintains near-perfect recall over context windows up to millions of tokens.
Why it matters
Pushed usable context length into the millions, enabling whole-codebase, long-video, and book-length reasoning.