Gemini 1.5
By Google DeepMind
5 is a family of multimodal large language models from Google DeepMind, released starting February 2024 in Pro and Flash variants, notable for introducing a breakthrough long context window of up to one million tokens (with an experimental…
Definition
Gemini 1.5 is a family of multimodal large language models from Google DeepMind, released starting February 2024 in Pro and Flash variants, notable for introducing a breakthrough long context window of up to one million tokens (with an experimental two-million-token version) using a mixture-of-experts architecture.
Overview
Gemini 1.5 followed Google DeepMind's original Gemini 1.0 family (Ultra, Pro, Nano) and represented a major architectural and capability upgrade, most prominently in context length. Gemini 1.5 Pro launched supporting a context window of up to 1 million tokens in general availability, with an experimental 2-million-token version demonstrated by Google, at the time far exceeding the context windows offered by competing frontier models. This long-context capability enabled genuinely new use cases, such as asking questions across an entire multi-hour video, a large codebase, or hundreds of pages of documents in a single prompt without external retrieval infrastructure, and Google published strong 'needle in a haystack' evaluation results showing the model could reliably locate specific facts embedded deep within very long inputs. Architecturally, Gemini 1.5 uses a mixture-of-experts (MoE) transformer design, activating only a subset of the model's total parameters for any given input, which Google DeepMind stated improved training and serving efficiency compared to a dense model of comparable quality. Like other Gemini models, it is natively multimodal, trained from the start to process and reason across text, images, audio, and video within a unified architecture rather than bolting on modality-specific components after the fact. Google released Gemini 1.5 in two primary tiers: Gemini 1.5 Pro, the higher-capability model aimed at complex reasoning and long-context tasks, and Gemini 1.5 Flash, a smaller, faster, and cheaper variant optimized for high-volume, latency-sensitive applications while retaining a large context window. Both were made available through the Gemini API, Google AI Studio, Vertex AI, and integrated into consumer-facing Google products. Gemini 1.5 was subsequently succeeded by the Gemini 2.x model generations, which built on its long-context and multimodal foundations with further reasoning and efficiency improvements.
Key Features
- Breakthrough context window of up to 1 million tokens (experimentally up to 2 million)
- Mixture-of-experts (MoE) transformer architecture for training and serving efficiency
- Natively multimodal: unified processing of text, images, audio, and video
- Strong 'needle in a haystack' long-context retrieval performance
- Released in Pro (higher capability) and Flash (faster, cheaper) tiers
- Available via the Gemini API, Google AI Studio, and Vertex AI
- Successor to the original Gemini 1.0 family (Ultra, Pro, Nano)
- Enables whole-document, whole-codebase, or whole-video prompting without external retrieval
Use Cases
Alternatives
Frequently Asked Questions
From the Blog
Claude vs ChatGPT vs Gemini: Which Is Best?
A comprehensive guide to claude vs chatgpt vs gemini: which is best? — written for learners at every level.
Read More AI & TechnologyLarge Language Models (LLMs) Explained for Beginners
An LLM predicts the next piece of text, one token at a time — this guide explains how ChatGPT, Claude, and Gemini actually work.
Read More