Choose the most appropriate model automatically

未关闭

#494 创建于 2025年9月9日 0 条评论 0 个 reaction 已指派 0 人在 GitHub 查看

还没有人认领这个 Issue。

评估

难度: 5/5
预计耗时: 一周以上
新手友好度: 25/100

Issue 类型: 功能
描述清晰度: 需要澄清
活跃度: 停滞
技术栈: swift
领域: ai, desktop-dev

调研方向

该 issue 没有指定任何文件、测试或入口点。首先定位现有的模型选择和 prompt 上下文处理流程，然后确定如何评估语义分析、任务难度、上下文长度和定价。完成的标准是自动选择适当且经济高效的模型，而不要求开发者手动切换模型。

由索引模型根据 Issue 内容生成。

描述

Currently, developers need to select a model on their own, and the variables involved in this choice include:

The capabilities of the model itself
The model’s context length
The pricing of the model (unlimited free usage, limited usage with different prices)

My suggestion:

First, Copilot should perform semantic analysis on the user’s current input prompt to determine the appropriate context. For example, if the user has started a new topic, previous context can be discarded. If the user is referring to modifying previous content or correcting errors, then the prior context should be retained.

After that, Copilot can begin to assess the difficulty of the question. For instance, is it a simple UI change or a complete UI design? Is it about fixing an error highlighted by the editor, or about correcting a logical error in a previous answer, etc.?

Finally, Copilot can choose the most cost-effective model based on the context length and the difficulty of the problem.

The reason I want this feature is that I often switch to the highest tier, like Sonnet 4 or GPT-5, and forget to switch back to GPT-4.1. This leads to unnecessary consumption of advanced requests.

主要语言: Swift
星标: 6.3k
派生: 2k
平均合并: 3 天 21 小时
30 天内合并 PR: 2