Google DeepMind·Gemini

Gemini 1.5 Pro

Used an MoE architecture to extend usable context to 1M tokens while retaining cross-modal retrieval.

LegacyLanguageMultimodalWeights not released

MODEL NUMBER012

Specifications

Only published specifications are shown.

Architecture

MoE

Context

Input modalities

Text · Image · Audio · Video

Open weights

Status

Legacy

Used an MoE architecture to extend usable context to 1M tokens while retaining cross-modal retrieval.

Gemini 1.5 Pro moved million-token context from demonstration to developer preview, resetting expectations for how much a model could read at once.

PreviousGemini 1.0 Ultra— · — ctx

CurrentGemini 1.5 Pro— · 1M ctx

New capabilities Long context Multimodal

Long contextMultimodal

2024-02-15Preview

Gemini 1.5 Pro entered limited preview with a 1M context window.

Select a card to view the related model.