Context Window
The maximum number of tokens a large language model can process in a single inference pass, determining the range of historical information visible to the model. Content exceeding the context window limit is truncated, directly affecting the design of long-conversation systems and constraining how much prior dialogue or document content a model can reference.
Core Facts
The context window of a large language model determines the upper limit of information the model can process in a single inference.
早期GPT-3的上下文窗口仅有4K Token，而现代模型已扩展至128K甚至更长
An Agent's memory mechanism is typically divided into two categories: short-term memory and long-term memory, with long-term memory requiring external persistent storage such as vector databases, relational databases, and graph databases.
Timeline (last 90 days)
The context window refers to the maximum total number of tokens a model can process in a single request, including the input prompt and the model output.
The context window determines how much content a model can remember in a single task; when the amount of project code exceeds the context window, the AI may forget previous design decisions and produce contradictory code.
The fuller the context window, the more the model tends to forget earlier information, while inference latency and API call costs grow linearly with token count.
Memory length is limited by the large language model's context window; advanced paid models with larger context windows can support stronger memory capabilities.
The context windows of today's top models can hold approximately one million tokens, roughly equivalent to eight novels.
Context compaction is a class of techniques that distill existing conversation history into a more concise representation, typically by having the model generate a summary of earlier content to replace the original message list.
The 1 million token output limit refers to the ceiling on content the model can generate in a single response, not the context window (input) length.
The context window is the limit on the amount of text a large model can process at one time; early models had only a few thousand tokens, while the Claude series of models has expanded to hundreds of thousands or even higher magnitudes
When handling large codebases, context expands rapidly, and once the context window is exceeded, the agent experiences forgetting; it is recommended to externalize background information to tools like Notion for the agent to retrieve on demand.
When the context window is insufficient, tools can only process files in batches, and the ability to make cross-file associative inferences degrades significantly.
25 more timeline events
All Facts (20)
The context window of a large language model determines the upper limit of information the model can process in a single inference.
90%Verified早期GPT-3的上下文窗口仅有4K Token，而现代模型已扩展至128K甚至更长
90%VerifiedAn Agent's memory mechanism is typically divided into two categories: short-term memory and long-term memory, with long-term memory requiring external persistent storage such as vector databases, relational databases, and graph databases.
80%Verified研究表明当上下文过长时，模型对中间部分信息的注意力会显著下降，即"Lost in the Middle"现象
80%VerifiedLarge language models are inherently stateless, and remembering conversation history relies on concatenating the historical conversation text and feeding it into the model together.
80%Verified大语言模型在长时间对话中会因上下文窗口限制而丢失早期信息
80%VerifiedThe context window refers to the maximum amount of text a model can process in a single conversation; early models had only 4K tokens, while modern models have expanded to hundreds of thousands of tokens.
70%VerifiedWhen conversation history exceeds the context window limit, platforms use engineering methods such as truncation and summary compression to trim the context, a process that inevitably loses detail information.
70%VerifiedThe context window determines how much content the model can remember within a single conversation.
65%VerifiedThe context windows of current mainstream large language models have expanded to 128K or even longer.
65%VerifiedThe context windows of current mainstream large language models have expanded to hundreds of thousands or even millions of tokens, but under ultra-long contexts the model's attention weights on earlier information decay significantly (the "lost in the middle" problem).
65%VerifiedModern Agent systems can maintain a context window of up to tens of thousands of tokens within a single session and perform multi-step logical decomposition of complex experimental plans through chain-of-thought reasoning mechanisms
65%VerifiedContext management can control token consumption through strategies such as sliding windows, summary compression, or hierarchical memory.
65%VerifiedAI Agents perform a large number of exploratory operations when handling ambiguous problems, causing token consumption to grow exponentially.
65%VerifiedCommon context pruning strategies include sliding window (retaining the most recent N messages), summary compression (summarizing old messages into a summary), and importance filtering (prioritizing the retention of tool call results).
65%Unverified大模型的API调用通常按Token数量计费，且每个模型都有上下文窗口（Context Window）的Token数量限制
95%Unverified大语言模型在处理任务时，会将读取到的文件内容纳入上下文窗口进行推理，一旦读取包含密钥的.env文件，密钥字符串可能被记录到日志或发送到模型服务商的API端点
90%Unverified大语言模型的所有记忆都存储在当前的上下文窗口（Context Window）中，开启新Thread后这些信息全部清空
90%Unverified在多Agent群聊中，成员数量过多会导致上下文混乱，降低协作效率，因为每增加一个成员会更快填满可用Token空间
85%Unverified每个 Claude 实例都有独立的上下文窗口（Context Window），子 Agent 从全新的上下文窗口启动，不携带主 Agent 的历史
85%