Long Context Window

The Long Context Window is a massive expansion of an LLM's working memory, enabling single-session processing of millions of tokens for complex, large-scale tasks.

This technology directly addresses the AI 'memory' problem: it dramatically increases the number of tokens (input and output) a model processes simultaneously. Current state-of-the-art models, like Gemini 1.5 Pro, handle up to 2 million tokens, which is roughly 5,000 pages of text. This capacity unlocks critical enterprise use cases: ingesting entire codebases for cross-file analysis, analyzing thousands of pages of legal contracts, or enabling advanced in-context learning for rare languages (e.g., Kalamang translation). While 1 million tokens is a current benchmark (Claude Sonnet 4), research is pushing limits, with models like Magic.dev's LTM-2-Mini testing up to 100 million tokens: the goal is full, persistent situational awareness for the most complex reasoning and retrieval tasks.

https://cloud.google.com/blog/products/ai-machine-learning/what-is-long-context-and-why-does-it-matter-for-ai