Models and thinking depth模型与思考深度

Pick one from the Model menu on the run form; the Thinking menu beside it then offers only the levels that model accepts. The menu starts on the server's default, or on the choice you last used in this browser.

A choice marked not configured here is listed but disabled in the menu: this server holds no credential for that provider, and a run started on it could not begin.

在分析表单的“模型”菜单中选择一项；旁边的“思考深度”菜单只会列出该模型支持的档位。菜单默认停在服务器的默认选项，或者你在这个浏览器上最近一次的选择。

标注本站未配置的选项会列在菜单中，但无法选择：本站没有该服务商的调用凭证，用它发起的分析根本无法启动。

Every choice names two models, because a run gives them different jobs:

• The quick model does most of the work: every analyst and their many data-tool calls, both researchers, the trader and the three risk analysts.
• The deep model is used by exactly two seats — the research manager and the portfolio manager — which is where the rulings are made.

That is why a choice can pair a stronger deep model with a lighter quick one: the many tool calls are what make a run expensive, and the decision you read keeps the more capable model. Follow-up questions on a finished report are answered by a separate fast model working from the report's text, whichever models wrote it.

每个选项都包含两个模型，因为一次分析会给它们分配不同的工作：

• 快速模型承担大部分工作：所有分析师及其大量的数据工具调用、两位研究员、交易员和三位风险分析师。
• 深度模型只用于两个席位——研究主管和投资组合经理——也就是做出裁决的地方。

所以一个选项可以把较强的深度模型和较轻的快速模型搭配在一起：让一次分析变贵的，正是大量的工具调用，而你最终读到的决策仍然由能力更强的模型做出。对已完成报告的追问由另一个独立的快速模型根据报告原文作答，与报告由哪个模型写成无关。

Thinking depth sets how much the model reasons before it answers — depth, not length. The levels are Minimal, Low, Medium and High, and each model accepts its own subset, shown in the table above.

• The Thinking menu only ever offers the chosen model's levels, and switching models resets it to that model's default.
• That strictness is deliberate: offering a level the provider silently rewrites would misdescribe the run, and some providers reject an unsupported level outright — minutes into a run you have paid for.
• A model with no thinking setting shows Not applicable, and nothing is sent.

思考深度决定模型在作答之前推理多少——是深度，而不是篇幅。档位依次为最小、低、中、高，每个模型只支持其中一部分，见上表。

• “思考深度”菜单只会列出当前模型支持的档位；切换模型时，会重置为该模型的默认档位。
• 这样严格是有意为之：如果提供一个会被服务商悄悄改写的档位，就等于错误描述了这次分析；而有些服务商会直接拒绝不支持的档位——那时分析已经开始好几分钟，次数也已经付出。
• 没有思考深度设置的模型会显示不适用，也不会发送这一参数。

The provider, the models and the thinking level are fixed when you press Run; a queued run uses exactly what you chose, even if the menu changes meanwhile. Every finished report names them — on the report page, in the completion email, on a shared link and on the public sample reports — so a run on the cheapest tier cannot pass for one on the most capable.

• Fallback. If a Gemini model runs out of its daily capacity mid-run, the run continues on a fallback Gemini model instead of failing, and the report is labelled ran on fallback model with the models it used — the PDF included.
• Stale choices. If your browser remembered a choice that no longer exists, the run uses the server default and records that instead.

服务商、模型和思考深度在你点击运行时就已确定；排队中的任务会严格按照你的选择运行，即使菜单在此期间有所变化。每份完成的报告都会注明这些信息——在报告页面、完成通知邮件、分享链接和公开的报告样例中——所以低价档位的分析，不会被误认为最高档位。

• 降级。如果某个 Gemini 模型在运行中途用完了当日额度，分析会换用备用的 Gemini 模型继续，而不是直接失败；报告会标注已降级模型并列出实际使用的模型，PDF 中也一样。
• 过期的选择。如果浏览器记住的选项已经不存在，分析会使用服务器默认模型，并如实记录。

Nothing extra. Every analysis costs one run — a free one or a credit — whichever model and thinking level you choose, because the allowances count runs, not tokens. A choice that is not available here cannot be picked, and a request for one is refused without costing anything. See Pricing and credits.

不额外收费。无论选择哪个模型、哪种思考深度，每次分析都只算一次——免费次数或已购次数——因为额度按分析次数计算，而不是按 token 计算。本站不可用的选项无法选中；即使直接请求，也会被拒绝，不扣任何次数。详见价格与次数。