长用 · 让时间和花费都值得Sustain · Make the Trade-Off Worthwhile.
用一周记录区分模型、等待和额度问题。看清内含用量、credits、API 与速度档的账, 留下真正减少返工的习惯。Use a week of records to distinguish model quality, waiting, and capacity. Separate included usage, credits, API billing, and speed modes, and keep habits that reduce rework.
用了一个星期, 你可能生成了十几份文件, 也可能只留下两份。但留下的那两份, 如果确实解决了每月对账和每周备课的问题, 就足够成为继续使用的理由。这一章把模型、额度和习惯放在一起:先看哪一步耗费你的时间, 再决定要不要花更多钱。After a week you may have generated a dozen files and kept only two. If those two genuinely help with monthly records or weekly preparation, that is enough reason to continue. This chapter brings models, usage, and habits together: identify where your time goes before deciding to spend more.
— I
三个账本, 不要混着算Keep Three Ledgers Separate.
截至 2026-10-09, 官方将 Work 和 Codex 列为 Free、Go、Plus、Pro、Business、Edu、Enterprise 计划所含体验。这里的「包含」不表示无限量、所有模型或所有入口都开放。Work 与 Codex 共享用量, 切换界面不会再生出一份额度。1注 1Note 1OpenAI · Pricing。Work / Codex 包含关系、模型与速度费率;截至 2026-10-09。本章为快照, 不是购买承诺。OpenAI · Pricing. Work / Codex inclusion, models, and speed rates; checked 2026-10-09. This chapter is a snapshot, not a purchase guarantee.As of 2026-10-09, official pricing includes Work and Codex with Free, Go, Plus, Pro, Business, Edu, and Enterprise plans. Inclusion does not imply unlimited usage or every model and surface. Work and Codex share usage; changing the interface does not create another allowance.1注 1Note 1OpenAI · Pricing。Work / Codex 包含关系、模型与速度费率;截至 2026-10-09。本章为快照, 不是购买承诺。OpenAI · Pricing. Work / Codex inclusion, models, and speed rates; checked 2026-10-09. This chapter is a snapshot, not a purchase guarantee.
账本Ledger
它代表什么What it means
别这样推断Do not infer
Included usage
计划内含的使用额度, 有适用限制和重置条件Usage included with a plan, subject to its limits and resets
从月费除以 token 单价反算可用次数Task counts by dividing subscription price by token price
Usage credits
符合条件功能超出内含量后的扩展用量Additional eligible usage beyond included limits
充值就升级套餐、解锁所有模型A top-up upgrades the plan or unlocks all models
API billing
API key 对应的独立开发者账单A separate developer bill associated with an API key
ChatGPT 订阅或 credits 自动抵扣 APIChatGPT subscriptions or credits pay API charges
个人计划先使用内含额度, 达到适用限制后才消耗 credits。Plus / Pro 的符合条件 agentic 功能可能共享同一个池。Free / Go 目前只有一部分账号可购买;看不到 Add credits 就不能假定自己能买, 已有余额与新购买资格也不是一回事。2注 2Note 2OpenAI Help · Using Credits for Flexible Usage in ChatGPT (Personal plans)。先用内含额度, Free / Go 购买灰度与自动充值条件;截至 2026-10-09。OpenAI Help · Using Credits for Flexible Usage in ChatGPT (Personal plans). Included usage is used first; Free / Go purchasing and automatic reload are conditional; checked 2026-10-09.普通 Chat、Work 任务与语音还有各自适用规则, 不要用一次聊天的限额替另一种活动估价。3注 3Note 3OpenAI Help · Using Codex with your ChatGPT plan。用量看板、共享池、语音和任务消耗的边界;截至 2026-10-09。OpenAI Help · Using Codex with your ChatGPT plan. Usage dashboards, shared pools, and voice/task boundaries; checked 2026-10-09.Personal plans use included usage before credits after applicable limits. Eligible agentic features on Plus / Pro can share a pool. Only some Free / Go accounts can currently purchase credits; absence of Add credits means you should not assume purchasing access, and an existing balance is distinct from eligibility to buy more.2注 2Note 2OpenAI Help · Using Credits for Flexible Usage in ChatGPT (Personal plans)。先用内含额度, Free / Go 购买灰度与自动充值条件;截至 2026-10-09。OpenAI Help · Using Credits for Flexible Usage in ChatGPT (Personal plans). Included usage is used first; Free / Go purchasing and automatic reload are conditional; checked 2026-10-09. Regular Chat, Work tasks, and voice have their own applicable rules; one activity's limit is not a quote for another.3注 3Note 3OpenAI Help · Using Codex with your ChatGPT plan。用量看板、共享池、语音和任务消耗的边界;截至 2026-10-09。OpenAI Help · Using Codex with your ChatGPT plan. Usage dashboards, shared pools, and voice/task boundaries; checked 2026-10-09.
— II
先找失败原因, 再换模型Identify the Failure Before Changing Models.
缺原始流水, 换大模型也算不出漏掉的扣费。格式不喜欢, 先给一段你认可的例子。任务太散, 缩到一份成品。只有材料齐全、标准明确后, 它仍不能处理关键约束或复杂推理, 才比较更强模型或更高推理力度。速度档主要处理等待, 不能替代质量验收。A larger model cannot recover missing transactions without source records. If the format is wrong, show an example you like. If the task is scattered, narrow it to one deliverable. Compare a stronger model or higher reasoning effort when complete inputs and clear criteria still leave important constraints unresolved. A speed tier addresses waiting, not acceptance.
你遇到的状况What happened
先试的动作First adjustment
答案泛泛, 不像自己的事Generic output
补用途、资料范围与具体例子Add purpose, source scope, and a concrete example
漏约束或反复推错Missed constraints or repeated reasoning errors
列验收清单;必要时提高模型/力度List acceptance checks; then consider a stronger model/effort
逻辑已对, 等待影响正在做的工作Correct work, but waiting blocks you
比较 Standard 与可用 Fast 的实测时间和用量Compare measured time and usage for Standard and available Fast
限额太快用完Allowance depletes quickly
减少重复上下文、无关工具、反复重做和不必要的子 AgentReduce repeated context, irrelevant tools, reruns, and unnecessary subagents
本书技术示例采用截至 2026-10-09 的 GPT-6 Luna 做窄探索、GPT-6.1 Sol 做实现与复核。它们是便于复现的选择, 不是质量排名。自己的账号未开放时, 使用可用模型与其支持的力度。把模型名当成可替换参数, 把验收材料留住, 才能在下次模型变动后重新比较。Technical examples use GPT-6 Luna for narrow exploration and GPT-6.1 Sol for implementation and review as of 2026-10-09. These are reproducibility choices, not a ranking. Use an available model and supported effort if your account lacks them. Treat model names as replaceable parameters and retain acceptance cases for future comparisons.
— III
看懂速度倍率与 token 费率Read Speed Multipliers and Token Rates Correctly.
下面是截至 2026-10-09 的消耗水平比, 每格分母都是同一模型的 Standard, 不是「速度提升百分比」或相对旧价的变化率。Ultra 推理档与 Ultrafast 速度档也要分开。1注 1Note 1OpenAI · Pricing。Work / Codex 包含关系、模型与速度费率;截至 2026-10-09。本章为快照, 不是购买承诺。OpenAI · Pricing. Work / Codex inclusion, models, and speed rates; checked 2026-10-09. This chapter is a snapshot, not a purchase guarantee.These are consumption-level ratios as of 2026-10-09. Every denominator is Standard for the same model, not speed improvement or a change from an older price. Ultra reasoning and Ultrafast speed are separate choices.1注 1Note 1OpenAI · Pricing。Work / Codex 包含关系、模型与速度费率;截至 2026-10-09。本章为快照, 不是购买承诺。OpenAI · Pricing. Work / Codex inclusion, models, and speed rates; checked 2026-10-09. This chapter is a snapshot, not a purchase guarantee.
速度档Speed tier
内含额度消耗 / 同模型 StandardIncluded usage / same-model Standard
购买 credits 消耗 / 同模型 StandardPurchased credits / same-model Standard
Fast
2.5×
2×
GPT-6 Astra / GPT-6.1 Sol Ultrafast
8×
6×
这些档位须模型与账号支持, 并非人人能选。不能把 Fast 写成统一 2.5 倍 credits, 也不能保证任务完成得快 2.5 倍。工具等待、网页响应、文件体积和返工都影响实际耗时。没有急用, 先以 Standard 完成一遍可验收的工作。These tiers require model and account support. Fast is not universally 2.5× credits, nor does it guarantee a 2.5× faster task. Tool waits, websites, file size, and rework all affect elapsed time. Unless urgency matters, first complete a verifiable run at Standard.以下只列本书用到的少量模型。单位为 Standard 下每 1,000,000 个对应类型 token 的 credits, 截至 2026-10-09;不是美元 API 价, 也不是订阅内含额度的兑换表。缓存输入与未缓存输入分别计一次, 不要把缓存再加进未缓存输入。The compact table covers models used in this book. Units are credits per 1,000,000 tokens of the indicated type at Standard speed, as of 2026-10-09. These are neither USD API prices nor a conversion table for included usage. Count cached and uncached input separately, without counting cached input twice.
一个任务的模型 token 账;主线程与所有子线程相加Model-token accounting for one task; sum parent and all children
人为算例:Sol 用 20,000 未缓存输入、80,000 缓存输入和 4,000 输出 token, Standard 为 1 + 0.2 + 1 = 2.2 credits;若全程适用 Fast 的购买 credits 费率, 同一 token 账为 4.4。这个算例不含另计的语音、图像或其他工具, 不代表真实任务会产生这些 token。完整任务以实际用量记录为准。Synthetic example: Sol uses 20,000 uncached input, 80,000 cached input, and 4,000 output tokens. Standard costs 1 + 0.2 + 1 = 2.2 credits. If the entire run qualifies for purchased-credit Fast pricing, the same token ledger costs 4.4. This excludes separately billed voice, images, or other tools and does not predict actual token production. Use actual usage records for the complete task.工作账号要先问管理员使用 credit 账本还是 USD 合同。旧 Codex rate-card 地址现跳转 Business / Enterprise / Edu 的表, 不应直接拿它替个人计划报价。4注 4Note 4OpenAI Help · ChatGPT Rate Card (Business, Enterprise/Edu credit-based pricing)。旧 Codex rate-card 地址现跳转此页, 适用企业 credit 账本, 不能冒充个人计划或 USD 合同;截至 2026-10-09。OpenAI Help · ChatGPT Rate Card (Business, Enterprise/Edu credit-based pricing). The old Codex rate-card URL redirects here; this workspace credit card is not a personal-plan or Enterprise USD contract quote; checked 2026-10-09.在支持的桌面用量界面查看单个 Work / Codex 聊天消耗;它不包括普通 Chat, 估算金额也不是发票。3注 3Note 3OpenAI Help · Using Codex with your ChatGPT plan。用量看板、共享池、语音和任务消耗的边界;截至 2026-10-09。OpenAI Help · Using Codex with your ChatGPT plan. Usage dashboards, shared pools, and voice/task boundaries; checked 2026-10-09.For a work account, ask whether billing uses credits or a USD agreement. The old Codex rate-card URL now redirects to the Business / Enterprise / Edu card; do not treat it as a personal-plan quote.4注 4Note 4OpenAI Help · ChatGPT Rate Card (Business, Enterprise/Edu credit-based pricing)。旧 Codex rate-card 地址现跳转此页, 适用企业 credit 账本, 不能冒充个人计划或 USD 合同;截至 2026-10-09。OpenAI Help · ChatGPT Rate Card (Business, Enterprise/Edu credit-based pricing). The old Codex rate-card URL redirects here; this workspace credit card is not a personal-plan or Enterprise USD contract quote; checked 2026-10-09. Supported desktop usage views can show individual Work / Codex chats, excluding regular Chat; estimated dollar values are not invoices.3注 3Note 3OpenAI Help · Using Codex with your ChatGPT plan。用量看板、共享池、语音和任务消耗的边界;截至 2026-10-09。OpenAI Help · Using Codex with your ChatGPT plan. Usage dashboards, shared pools, and voice/task boundaries; checked 2026-10-09.
— IV
花钱前, 给自动化设停止条件Set Stop Conditions Before Spending.
先记录一次人工发起的消耗, 再决定让它每周跑。把连续失败、资料缺失、来源改版、重复通知列为暂停原因。需要提高花费或更大权限时回来问你, 不让一个错误定时重复放大。取消不用的日程, 关闭不需要的连接, 比换一个便宜模型更容易消除浪费。Measure one manually initiated run before scheduling it weekly. Pause for repeated failure, missing materials, changed sources, or duplicate notifications. Require a return to you before expanding spend or access. Remove unused schedules and connections; this can eliminate waste more directly than changing to a cheaper model.自动充值只在符合条件的账号上提供。启用时若余额已经低于门槛, 可能立刻购买;自动充值的月上限不包含另做的一次性购买。当前个人 credits 有购买后 12 个月有效期, 不可转让, 通常不退款但法律要求除外, 截至 2026-10-09。先核对结算页, 不为了本书练习打开自动充值。2注 2Note 2OpenAI Help · Using Credits for Flexible Usage in ChatGPT (Personal plans)。先用内含额度, Free / Go 购买灰度与自动充值条件;截至 2026-10-09。OpenAI Help · Using Credits for Flexible Usage in ChatGPT (Personal plans). Included usage is used first; Free / Go purchasing and automatic reload are conditional; checked 2026-10-09.Automatic reload is conditional. Enabling it with a balance below the threshold may trigger an immediate purchase, and its monthly cap does not include separate one-time purchases. Personal credits currently expire 12 months after purchase, are non-transferable, and generally non-refundable except as required by law, as of 2026-10-09. Check checkout terms; these exercises do not require automatic reload.2注 2Note 2OpenAI Help · Using Credits for Flexible Usage in ChatGPT (Personal plans)。先用内含额度, Free / Go 购买灰度与自动充值条件;截至 2026-10-09。OpenAI Help · Using Credits for Flexible Usage in ChatGPT (Personal plans). Included usage is used first; Free / Go purchasing and automatic reload are conditional; checked 2026-10-09.
— V
留下三条真正有用的习惯Keep Three Habits That Actually Help.
一周记录四段时间:准备、等待、修订、验收。只看等待时间会漏掉你整理材料和纠错的劳动。再记本次有没有可用成品、下次能否复用、用了哪个模型和入口、看板显示什么用量。没有数字就写未记录, 不让模型替你编节时率。Record preparation, waiting, revision, and acceptance time for a week. Measuring only waiting hides your work organizing sources and fixing errors. Also note whether the result was useful, reusable, which model and surface you used, and what usage was displayed. Mark missing numbers unrecorded instead of asking a model to invent savings.
提示词PromptChat / Work · Weekly review
根据我提供的一周记录, 帮我选出值得继续的三类任务。分别看可用成果、准备时间、等待、返工、验收和实际花费。缺失数据标未记录, 不推测节时率。建议一项停止、一项简化、一项保留的流程;不要自动修改日程或购买订阅。Use my weekly log to choose three task types worth continuing. Consider useful results, preparation, waiting, revision, acceptance, and actual cost. Mark missing data unrecorded and do not infer time-saving percentages. Suggest one workflow to stop, one to simplify, and one to keep. Do not change schedules or buy a subscription.
本周三项记录:口语练习完成4次, 每次15分钟, 没记用量;订阅对账准备10分钟、等待3分钟、纠错8分钟、核对12分钟, 留下可编辑表格;旅行计划改了5轮, 仍没确定同行者预算, 没有可用日程。请指出下一周怎么调整。不要给我一个虚构的效率提升百分比。This week: four 15-minute speaking practices, usage unrecorded; subscription reconciliation took 10 minutes preparation, 3 waiting, 8 correction, 12 verification, and produced an editable sheet; travel planning took five revisions but companions' budgets remain unknown and no usable itinerary exists. Suggest next week's changes without inventing an efficiency percentage.
这个例子里, 口语练习可以保留, 但还要用脱离 AI 的自测确认学习效果;对账值得把分类规则保存下来;旅行计划应暂停重写, 先问清同行者预算。这些决定来自记录和任务条件, 与用了几项新功能无关。In this example, keep speaking practice while checking learning without AI, save the reconciliation rules for reuse, and pause itinerary rewrites until companions' budgets are known. These decisions follow the record and task conditions, not the number of new features used.这本书最后留给你的, 可以很小:一份每月能对上的账、一套你愿意继续的练习、一个不再需要反复手工操作的小工具。把成果存好, 把来源和验收保留, 把不合适的任务停掉。产品下一次变化时, 你仍然知道自己需要什么。What you keep from this book can be small: records you can reconcile monthly, practice you want to continue, and a tool that removes a repetitive chore. Save the results, preserve sources and checks, and stop unsuitable tasks. When the product changes again, you will still know what you need.
留下对生活有用的成果, 也留下自己的判断.
Keep the results that help your life, and keep your own judgment.
SUSTAIN · USE WHAT HELPS
— 讨论Discussion
讨论Discussion.
评论区初始化中…Initializing comments…
01 / 01
没有匹配结果No matches.
换个关键词,或按 Esc 回到页面Try another keyword, or press Esc to return