From the basic series: context window. A language model can only look at what's inside its window while writing an answer; Today, in the most powerful models, this window is around one million tokens. So, does the model use everything that enters the window equally well? We describe what the window is made of, the cost of its length, and three metrics: Lost in the Middle's U curve (success drops when the answer is in the middle), RULER's declared and effective length difference, NoLiMa's no-word-match pin. Then, we show the phenomenon that Anthropic calls "context rot" and its solutions (putting the question at the end, not giving unnecessary tokens, compression, sub-agents) on the resources' own pages. Chapters 0:00 Opening: middle sentence 0:26 What is a context window? 0:54 What goes into the window? 1:27 How big is the window? 1:53 The price of size: n² and price 2:44 Lost in the Middle: U curve 3:29 RULER: declared and effective length 4:05 NoLiMa: without word matching 4:46 Context rot 5:06 What is done: question ends 5:20 Agents: compression and sub-agents 5:56 Back to the top document Note: Window sizes and prices are as of October 30, 2026. This video is not affiliated with Anthropic, Google, OpenAI, NVIDIA, Adobe or the authors of the articles. Sources Anthropic, Context windows: https://platform.claude.com/docs/en/build-with-claude/context-windows Anthropic, Effective context engineering for AI agents: https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents Anthropic, Models overview (1M token ≈ 555 thousand words): https://platform.claude.com/docs/en/models/overview Google, Long context (Gemini API): https://ai.google.dev/gemini-api/docs/long-context Google, Gemini 4 Argon model card: https://ai.google.dev/gemini-api/docs/models/gemini-4-argon OpenAI, GPT-5.6 Left (context window and 272K price rule): https://developers.openai.com/api/docs/models/gpt-5.6 Liu et al., “Lost in the Middle: How Language Models Use Long Contexts” (TACL): https://arxiv.org/abs/2307.03172 Hsieh et al., “RULER: What's the Real Context Size of Your Long-Context Language Models?” (NVIDIA, COLM 2024): https://arxiv.org/abs/2404.06654 Modarressi et al., "NoLiMa: Long-Context Evaluation Beyond Literal Matching" (Adobe Research, LMU Munich): https://arxiv.org/abs/2502.05167
The information provided is not trading advice. kdj.com does not assume any responsibility for any investments made based on the information provided in this article. Cryptocurrencies are highly volatile and it is highly recommended that you invest with caution after thorough research!
If you believe that the content used on this website infringes your copyright, please contact us immediately (info@kdj.com) and we will delete it promptly.