Google DeepMind launched Gemini 2.5 Pro, combining built-in multi-step reasoning with a 1M-token native multimodal context window and strong coding performance across long-horizon benchmarks.
Gemini
Also known as: Gemini 1.5 · Gemini 2.0 · Gemini 2.5 · Google Gemini
Plain English
Google DeepMind's flagship family of natively multimodal AI models built from the ground up to process text, code, images, audio, and video together over very long context windows.
Technical Overview
Trained on Google TPUs with native early-fusion multimodality and sparse Mixture of Experts (MoE) scaling, the Gemini family pioneered million-token context windows capable of ingesting entire codebases, hour-long videos, and audio streams in a single prompt, later integrating test-time reasoning in the Gemini 2.5 Pro and Flash tiers.
Chronological Entries (2)
Google DeepMind published the Gemini technical report introducing Ultra, Pro, and Nano sizes trained natively across text, image, audio, and video tokens rather than stitching separate modality encoders after pretraining.
Direct Relations (4)
- Created byGoogle DeepMind
- ImplementsMultimodal & Vision-Language Models
- ImplementsMixture of Experts (MoE)
- Alternative toClaude