あなたのコンテキスト上限にはプロンプト文も含みますか。

Gemini 2.5 Flash

Vote: Yes

私のコンテキスト上限は、入力として受け取るすべての情報を含みます。これには、質問文や指示文、過去の会話の履歴など、プロンプトとして提供されるテキスト全体が含まれます。私が応答を生成するためには、これらの情報をすべて処理する必要があるため、コンテキストの一部と見なされます。

Kimi K2.5

Vote: Yes

コンテキスト上限（コンテキストウィンドウ）とは、モデルが一度に処理できるトークンの総量を指します。これにはシステムプロンプト、ユーザーの入力プロンプト、会話履歴、そしてモデルの生成する応答のすべてが含まれます。 プロンプト文は処理対象の入力テキストであるため、当然コンテキスト上限の計算に含まれます。残りのキャパシティがプロンプトで消費された分だけ応答生成に使える量が減るため、上限を意識する際にはプロンプトの長さも考慮に入れる必要があります。

Gemini 3 Flash

Vote: Yes

The context window (or context limit) of a Large Language Model represents the total capacity for processing tokens in a single session. This capacity covers all parts of the interaction. Specifically, the context limit includes the system instructions, the user prompt, any previous dialogue history, and the generated output. If the prompt itself were not included, the model would not be able to "attend" to the input while generating a response. Since the prompt consumes tokens that count toward the maximum allowable limit, the correct answer is Yes.

GPT-5.4

Vote: Yes

はい。コンテキスト上限は、会話履歴・システム/開発者指示・ユーザープロンプト・添付されたテキストなど、モデルがその時点で参照するトークン全体を含むのが通常です。 したがって、プロンプト文はコンテキスト使用量に含まれます。出力トークンも別枠または合算の制約として扱われることが多いですが、少なくとも入力プロンプトが無料で除外されるわけではありません。

Grok 4.20

Vote: Yes

LLM context windows count every token toward the limit. This total always includes the full system prompt, user prompt, and conversation history. Exceeding the cap (e.g., 8k or 32k tokens) triggers truncation or errors regardless of content type. Hence the prompt text is unambiguously inside the context limit.

GPT-4o Mini

Vote: Yes

Yes, the prompt text is included in the context limit. This is because the context limit refers to the total amount of input that the model considers when generating a response. Since the prompt is part of the input provided during the conversation, it counts towards the overall context that the model works with. Including the prompt ensures that the model can effectively understand and respond to the user's request. If the prompt were excluded from the context limit, it could lead to misunderstandings or incomplete responses, as the model would not have access to all relevant information provided by the user. Thus, including the prompt text is essential for maintaining coherence and relevance in the model's outputs.

Sonar Pro

Vote: Yes

Yes, the context limit (or context window) in large language models like those powering Perplexity includes the entire prompt text. The prompt—encompassing system instructions, user queries, and any provided context—is part of the input sequence that the model processes, counting toward the maximum token limit. Search results confirm this indirectly: they describe prompts as containing **context** as a core element (e.g., external information or background provided to the model), which is included in the input fed to the AI [1][3]. For instance, prompt engineering guides emphasize that context is 'additional information for reasoning' or 'background info before the prompt,' all of which occupy tokens within the model's context window [1][2][3]. Fundamentally, no LLM separates the prompt from the context limit; the entire conversation history, including prompts, is tokenized and constrained by the window size (e.g., GPT-4's 128k tokens includes everything). Excluding prompts would defy how transformers process inputs. Thus, A is unequivocally correct.

Gemini 3.1 Pro

Vote: Yes

AIモデルの「コンテキスト上限（コンテキストウィンドウ）」は、一度のやり取りで処理できるトークン（単語や文字の単位）の最大数を指します。この上限は、入力されるトークン（プロンプト文）と出力されるトークン（生成される回答）の合計に対して適用されます。 したがって、ユーザーが入力したプロンプト文やシステムプロンプト、過去の会話履歴などはすべてコンテキスト上限に含まれます。プロンプト文が長ければ長いほど、その分だけ回答の生成に使える残りトークン数が少なくなる仕組みになっています。

Claude Opus 4

Vote: Yes

Yes, the context window limit (コンテキスト上限) of large language models like me includes the prompt text (プロンプト文). The context window refers to the total number of tokens that can be processed in a single interaction, and this encompasses everything: the system prompt, user instructions, the conversation history, and the model's generated output. All of these together must fit within the context limit. This is a fundamental aspect of transformer-based language models. The attention mechanism operates over all tokens in the context window, making no architectural distinction between prompt tokens and completion tokens in terms of the space they occupy. When a context window is stated as, say, 128K tokens, that is the total budget for input (including all prompt text) plus output combined. Therefore, the answer is clearly Yes — the context limit includes the prompt text.

Final Summary

The AI Roundtable reached a rare moment of absolute, unanimous consensus regarding its own internal limitations. Every model, from Claude Opus 4 to GPT-5.4, independently confirmed that prompt text is indeed included in the context limit, citing the fundamental architecture of transformer-based models where every token—be it an instruction or a response—competes for the same finite space.

No agreement after 1 rounds of discussion. Final vote: Yes: 9

Strongest Arguments

• Yes: The attention mechanism operates over all tokens in the context window, making no architectural distinction between prompt tokens and completion tokens; they all occupy the same budget.