Skip to content

Latest commit

 

History

History
767 lines (563 loc) · 35.9 KB

cli-command-reference.md

File metadata and controls

767 lines (563 loc) · 35.9 KB

CLI 指令与输出参考

状态:0.1.0 已实现,基于 TypeScript 与 Commander;npm 包名为 real-api-pricing,维护者发布后可用 npm install -g real-api-pricing 安装,当前可从仓库构建或安装本地 tarball。默认自动联网,失败时依次使用有效缓存、内置快照,正常回退不增加选项或网络警告。

下列数值与 Source: bundled 示例来自 2026-10-01 的历史验证;show/compare 的英文证据节选使用 2026-10-02 快照已有的译文。表格按字段整理,较长的详情只列节选;帮助采用当前英文输出。

同步主仓库 main 的 9be0bb9 后,当前内置快照为 2026-10-03,与本次验证的线上文件一致。该快照重新完成 Chromium 61 个场景、累计 5,176 行,以及英文详情、compare、CSV 和移动端检查,均通过。Droid 月额度已按上游调整,历史示例不代表当前价格或额度;实际日期与来源以每次查询的元信息为准。验证范围与 CSV 对账规则见 CLI 设计规范,安装见 CLI README。

命令名为 rap。默认表格、默认返回全部匹配记录;表头、套餐显示、帮助及错误信息固定使用英文。Model、Company、Channel、Plan 在每张价格记录表中独立成列,即使筛选后值相同也保留。Company 是模型开发公司,对应网站 vendors;Channel 是提供套餐的访问渠道。例如 Claude 由 Anthropic 开发,Droid Max 的渠道是 Factory。

下文标为 stdout 的 Markdown 表格是对应终端表格的字段示意,省略边框及换行;标为 stderr 的代码块是独立的结果说明,不混入表格、JSON 或 CSV。完整 Point ID 可以在窄终端换行,但不得截断。月额度的 B 表示十亿 token,真实单价单位为 USD/MTok;月费保留美元符号,原币 CNY 价格附在同一格。没有月额度或月费的值显示 —。Billing、Confidence 与网站使用相同英文显示名;JSON/CSV 保留原始标识和数值。排名与查询价格使用紧凑格式,show/compare 主价最多 6 位有效数字;query 分数最多 2 位小数,详情分数最多 4 位小数。排序使用原始精度。

1. 指令总览

指令 用途 默认次序
rap price 对齐网站真实单价排名,包含订阅和按量 API 真实单价升序
rap allowance 对齐网站月额度排名,只保留有月额度的订阅 月额度降序
rap list models 发现模型 ID、公司与覆盖记录数 模型 ID 升序
rap list companies 发现模型开发公司的标识与显示名 公司 ID 升序
rap list plans 发现套餐 ID、渠道及覆盖模型数 套餐 ID 升序
rap list channels 发现访问渠道及覆盖记录数 渠道名称升序
rap list boards 发现榜单、指标与独立快照日期 榜单 ID 升序
rap query 通用表格查询;自定义排序、评测配置与前沿 真实单价升序
rap show <point-id> 一条记录的价格、额度、采用依据与配置映射 指定记录
rap compare <point-id> <point-id>... 比较至少两条记录 保留输入顺序
rap info 查询依据、数量、负载与汇率口径 固定字段次序
rap --help / rap <command> --help 全局或子命令帮助 —
rap --version CLI 软件版本 —

全局数据选项为 --data <site.json>、--format table|json|csv,没有额外联网、离线或刷新选项。软件版本与数据快照日期分开;数据命令默认请求 https://realapipricing.com/data/site.json,利用 ETag/304 复用已校验缓存,失败时自动回退到缓存或内置快照。请求及响应体读取总超时为 3 秒。

数据保存在本地:仓库的 data/adopted.csv 与 derived/*.json 通过共享适配器分别生成网站的 web/public/data/site.json 和 CLI 的 cli/data/site.json;两者的快照内容相同。npm pack 将 CLI 快照随本地安装包携带,数据路径相对安装位置定位,在任意工作目录均可读取。--data <site.json> 可指定其他本地快照,显式文件覆盖会完全跳过联网和缓存。默认缓存位置及读取规则见 CLI README。

2. rap price:真实单价排名

对应网页 #lang=en&view=price&vendors=Anthropic。默认返回 Anthropic 的全部 29 条价格记录,包括第三方渠道和按量 API。下面显式限制为前三条:

rap price --company Anthropic --limit 3

stdout:

Rank Model Company Channel Plan USD/MTok Monthly tokens Monthly fee USD Billing Confidence Point ID
1 Claude Sonnet 5 Anthropic Anthropic Claude Pro $0.00309 6.472 B $20 Subscription Medium claude_pro::claude-sonnet-5
2 Claude Sonnet 5 Anthropic Anthropic Claude Max 20x (9/14+) $0.0051 39.25 B $200 Subscription Medium claude_max_20x::claude-sonnet-5
3 Claude Sonnet 5 Anthropic Anthropic Claude Max 5x (9/14+) $0.0051 19.625 B $100 Subscription Low claude_max_5x::claude-sonnet-5

stderr:

Snapshot: 2026-10-01 | Source: bundled | View: price
Total: 29 | Returned: 3 | Truncated: true
Order: USD/MTok ascending; ties by Point ID.

Company 与 Channel 可以同时筛选。以下查询仍然选择 Anthropic 模型,但只看 Factory 渠道:

rap price --company Anthropic --channel Factory --limit 3

stdout:

Rank Model Company Channel Plan USD/MTok Monthly tokens Monthly fee USD Billing Confidence Point ID
1 Claude Sonnet 5.5 Anthropic Factory Droid Max $0.01979 10.104 B $200 Subscription Medium droid_max::claude-sonnet-5.5
2 Claude Opus 5.5 Anthropic Factory Droid Max $0.03959 5.052 B $200 Subscription Medium droid_max::claude-opus-5.5
3 Claude Opus 4.8 Anthropic Factory Droid Max $0.04949 4.042 B $200 Subscription Medium droid_max::claude-opus-4.8

stderr:

Snapshot: 2026-10-01 | Source: bundled | View: price
Total: 7 | Returned: 3 | Truncated: true
Order: USD/MTok ascending; ties by Point ID.

price 始终一点一行,内部固定选最高匹配配置;--sort、--min-score、--scored-only、--frontier 和显式 --config 不适用于排名视图。直接传给 price 时 Commander 报未知参数;通过 query 的排名视图传入时会报告只适用于 table。

按量 API 仍能在 price 中查询,它的月额度和月费为不适用:

rap price --company Anthropic --model claude-opus-5.5 --billing metered

stdout:

Rank Model Company Channel Plan USD/MTok Monthly tokens Monthly fee USD Billing Confidence Point ID
1 Claude Opus 5.5 Anthropic Anthropic Claude Opus 5.5 API $0.419 — — Metered API High anthropic_opus55_api::claude-opus-5.5

stderr:

Snapshot: 2026-10-01 | Source: bundled | View: price
Total: 1 | Returned: 1 | Truncated: false
Order: USD/MTok ascending; ties by Point ID.

原币为 CNY 的套餐同时保留采用美元月费与原币价,不重新按汇率覆盖采用值:

rap price --plan kimi_allegretto_cn --model kimi-k3

stdout,<br> 表示同一单元格内换行:

Rank Model Company Channel Plan USD/MTok Monthly tokens Monthly fee USD Billing Confidence Point ID
1 Kimi K3 (v1) Kimi Kimi Kimi Allegretto $0.02688 1.451 B $39
¥199
Subscription Medium kimi_allegretto_cn::kimi-k3

stderr:

Snapshot: 2026-10-01 | Source: bundled | View: price
Total: 1 | Returned: 1 | Truncated: false
Order: USD/MTok ascending; ties by Point ID.

不计额度促销点没有月 token 分母,额度列显示 —;促销说明在该记录的续行:

rap price --plan devin_pro --model swe-2

stdout:

Rank Model Company Channel Plan USD/MTok Monthly tokens Monthly fee USD Billing Confidence Point ID
1 SWE-2 Devin Devin Devin Pro (promo until 10/31) ≈$0 — $20 Subscription Medium devin_pro::swe-2

该行之后的续行字段:

Point ID: devin_pro::swe-2
Promotion: promo until 2026-10-31, unmetered · $20 / mo

stderr:

Snapshot: 2026-10-01 | Source: bundled | View: price
Total: 1 | Returned: 1 | Truncated: false
Order: USD/MTok ascending; ties by Point ID.

3. rap allowance:月额度排名

对应网页 #lang=en&view=allowance&vendors=Anthropic。API 的月额度不适用,不进入这个排名;也不把未知额度解释为无限额度。Anthropic 筛选共有 24 条:

rap allowance --company Anthropic --limit 3

stdout:

Rank Model Company Channel Plan USD/MTok Monthly tokens Monthly fee USD Billing Confidence Point ID
1 Claude Sonnet 5 Anthropic Anthropic Claude Max 20x (9/14+) $0.0051 39.25 B $200 Subscription Medium claude_max_20x::claude-sonnet-5
2 Claude Opus 5.5 Anthropic Anthropic Claude Max 20x (9/14+) $0.00634 31.538 B $200 Subscription Medium claude_max_20x::claude-opus-5.5
3 Claude Sonnet 5 Anthropic Anthropic Claude Max 5x (9/14+) $0.0051 19.625 B $100 Subscription Low claude_max_5x::claude-sonnet-5

stderr:

Snapshot: 2026-10-01 | Source: bundled | View: allowance
Total: 24 | Returned: 3 | Truncated: true
Order: monthly tokens descending; ties by Point ID.
Fee band: all

月费档位与网站一致,0-30 包含 $30,30-100 不包含 $30、包含 $100,100-300 不包含 $100、包含 $300;all 也保留超过 $300 的套餐。

rap allowance --company Anthropic --fee-band 0-30 --limit 3

stdout:

Rank Model Company Channel Plan USD/MTok Monthly tokens Monthly fee USD Billing Confidence Point ID
1 Claude Sonnet 5 Anthropic Anthropic Claude Pro $0.00309 6.472 B $20 Subscription Medium claude_pro::claude-sonnet-5
2 Claude Opus 5.5 Anthropic Anthropic Claude Pro $0.00655 3.054 B $20 Subscription High claude_pro::claude-opus-5.5
3 Claude Opus 4.8 Anthropic Anthropic Claude Pro $0.00772 2.589 B $20 Subscription Low claude_pro::claude-opus-4.8

stderr:

Snapshot: 2026-10-01 | Source: bundled | View: allowance
Total: 6 | Returned: 3 | Truncated: true
Order: monthly tokens descending; ties by Point ID.
Fee band: 0-30

4. rap list:发现可查询资源

list 搜索资源自身的 ID 和显示名,返回资源汇总,不展开价格点。先按适用的 Company/Channel 条件确定统计范围,再按资源搜索和 limit 输出;计数为该范围内的记录数。支持情况如下,不相关参数直接报错:

资源 --search / --limit --company / --vendor --channel
models 支持 支持 支持
companies 支持 不适用 支持
plans 支持 支持 支持
channels 支持 支持 不适用
boards 支持 不适用 不适用

rap list models

用途:查准确的 Model ID;Company 单列,不混入模型名。

rap list models --company Anthropic --search claude-opus-5.5

stdout:

Model ID Model Company ID Company Plans Channels Points
claude-opus-5.5 Claude Opus 5.5 Anthropic Anthropic 7 3 7

stderr:

Snapshot: 2026-10-01 | Source: bundled | Resource: models
Total: 1 | Returned: 1 | Truncated: false

rap list companies

用途:查模型开发公司的原始标识和显示名。--company 精确筛选用 Company ID;例如 Muse 的显示名是 Meta,Cognition 的显示名是 Devin,仍保留原始 ID。

rap list companies --search Anthropic

stdout:

Company ID Company Models Plans Points
Anthropic Anthropic 7 12 29

stderr:

Snapshot: 2026-10-01 | Source: bundled | Resource: companies
Total: 1 | Returned: 1 | Truncated: false

rap list plans

用途:查套餐 ID 和访问渠道。一个套餐可以覆盖多个模型、多个开发公司;Models 是覆盖的不同模型数,不能把套餐视为只属于一个公司。

rap list plans --channel Factory --search droid_max

stdout:

Plan ID Plan Channel Companies (IDs) Models Points
droid_max Droid Max Factory Alibaba, Anthropic, DeepSeek, Google, Kimi, MiniMax, Mistral, OpenAI, SpaceXAI, Zhipu, other 27 27

stderr:

Snapshot: 2026-10-01 | Source: bundled | Resource: plans
Total: 1 | Returned: 1 | Truncated: false

rap list channels

用途:查访问渠道。以下输出统计 Factory 提供的 Anthropic 模型,不是 Factory 全部模型。

rap list channels --company Anthropic --search Factory

stdout:

Channel Models Plans Points
Factory 6 2 7

stderr:

Snapshot: 2026-10-01 | Source: bundled | Resource: channels
Total: 1 | Returned: 1 | Truncated: false

rap list boards

用途:查独立榜单的 ID、指标和快照。榜单日期不同于价格快照日期;分数不能跨榜单混比。

rap list boards --search aa_intelligence_index

stdout:

Board ID Board Metric Board snapshot Configurations
aa_intelligence_index Artificial Analysis Intelligence Index v4.3 Intelligence Index 2026-09-22 126

stderr:

Snapshot: 2026-10-01 | Source: bundled | Resource: boards
Total: 1 | Returned: 1 | Truncated: false

5. rap query:通用查询

默认 --view table,可筛选精确模型、公司、渠道、套餐、计费类型、置信度和数值范围,默认按真实单价升序。表格不要求有评测分数。

rap query --model claude-opus-5.5 --channel Factory

stdout:

Model Company Channel Plan USD/MTok Monthly tokens Monthly fee USD Billing Confidence Point ID
Claude Opus 5.5 Anthropic Factory Droid Max $0.03959 5.052 B $200 Subscription Medium droid_max::claude-opus-5.5
Claude Opus 5.5 Anthropic Factory Droid Pro $0.05236 0.382 B $20 Subscription Low droid_pro::claude-opus-5.5

stderr:

Snapshot: 2026-10-01 | Source: bundled | View: table
Total: 2 | Returned: 2 | Truncated: false
Order: price:asc

rap query --view price 与 rap price 的参数、数据、排序和输出相同;rap query --view allowance 与 rap allowance 相同,不另造一套排名:

rap query --view price --company Anthropic --limit 3
# stdout/stderr 与第 2 节第一个示例一致。

rap query --view allowance --company Anthropic --limit 3
# stdout/stderr 与第 3 节第一个示例一致。

评测查询只在 table 中展开配置。以下同一价格点输出三个配置,价格与额度重复是因为配置不同;不是三份可累加额度。

rap query --plan droid_max --model claude-opus-5.5 \
  --board aa_intelligence_index --config all --sort score:desc \
  --limit 3

stdout,价格区域:

Model Company Channel Plan USD/MTok Monthly tokens Monthly fee USD Billing Confidence Point ID
Claude Opus 5.5 Anthropic Factory Droid Max $0.03959 5.052 B $200 Subscription Medium droid_max::claude-opus-5.5
Claude Opus 5.5 Anthropic Factory Droid Max $0.03959 5.052 B $200 Subscription Medium droid_max::claude-opus-5.5
Claude Opus 5.5 Anthropic Factory Droid Max $0.03959 5.052 B $200 Subscription Medium droid_max::claude-opus-5.5

stdout,核心表下方的评测区域节选,按相同行序展示;完整输出另含 Point ID、Harness、Mode、Estimated、Self-reported、Best ties:

Score Effort Variant Configuration ID Mapping confidence Quota effort matched
57.62 Max Claude Opus 5.5 (max with fallback) aa_intelligence_index:6df46e4e9119d1e1 Medium unverified
55.99 xhigh Claude Opus 5.5 (xhigh with fallback) aa_intelligence_index:4c3fe13d36952d4f Medium unverified
53.58 High Claude Opus 5.5 (high with fallback) aa_intelligence_index:c4632b7e15588786 Medium unverified

stderr:

Snapshot: 2026-10-01 | Source: bundled | View: table
Board: aa_intelligence_index | Board snapshot: 2026-09-22 | Config: all
Total: 5 | Returned: 3 | Truncated: true
Order: score:desc

--config best 默认选所筛配置中的最高分;--frontier 须指定 --board,在完整筛选结果上计算后才应用 limit。无分记录默认保留,只有 --min-score、--scored-only 或 --frontier 才排除它们。JSON/CSV 的字段结构、前沿与缺失值规则见设计规范。

6. rap show:单条记录与采用依据

用途:用精确 Point ID 查看数据及证据。默认展示全部已映射榜单;下面用 --board 缩小到一个榜单,仍展示该榜单所有匹配配置。

rap show 'droid_max::claude-opus-5.5' --board aa_intelligence_index

stdout,核心数据:

Model Company Channel Plan USD/MTok Monthly tokens Monthly fee USD Billing Confidence Point ID
Claude Opus 5.5 Anthropic Factory Droid Max $0.0395883 5.052 B $200 Subscription Medium droid_max::claude-opus-5.5

stdout,原始数值与采用依据字段节选;完整表还包括 Promotion、List blended USD/MTok:

Field Value
Model ID claude-opus-5.5
Company ID Anthropic
Plan ID droid_max
Original monthly fee 200 USD
Local price —
Monthly tokens, raw 5,052,000,000
USD/MTok, raw 0.039588282
Workload Anthropic workload: 97% cache reads / 2.5% cache writes / 0.5% output
Data date 2026-09-29
Date kind sample
Date inherited from —
Unmetered false
Promotion until —
Plan generation
Official model prices Official API list (cached / in / out, per MTok): $0.2 / $4 / $20
Subscription price assumption Full use of the adopted allowance

stdout,Source 和 Decision note 优先显示网站的现有英文译文,缺译文时使用原文,允许按终端宽度换行。此处 note_en 与 decision_note_en 相同,因此不重复输出 Note:

Source:
Maintainer's local Factory Droid measurement: weekly quota (7-day rolling) used 1%→5% segment (xhigh) +35,598,616 tok and 5%@08:13→9%@12:40 segment (high, new session 47f9711d) +36,567,946 tok, total +72,166,562 tok all claude-opus-5-5 (auto 0 delta; glm-5.3-flash +50,317 belongs to the free Droid Core pool and is not counted); total breakdown in 793,017 / out 375,172 / cache_create 3,640,056 / cache_read 67,311,924 / thinking 46,393; droid-opus55-max-round1-2026-09-29.json; https://factory.ai/pricing Max $200/month

Decision note:
First Factory Droid point, normalized on the same basis as devin_max×opus-5.5: the combined 1→9 segment mix — cache read 93.27% / cache write 5.04% / input 1.10% / output 0.52% / thinking 0.06% (hit 93.82%) — deviates from the standard tier; at Opus 5.5 list prices cached $0.2 / 5m write $5 / in $4 / out $20 (thinking 46,393 unbilled per ruling; consistent with factoryCredits 16,935,482 ≈ worth÷$4×1.6), segment worth $42.34 ÷8%×4 weeks = $2,116.91/month list-worth ÷ Anthropic-tier blended price $0.419/MTok = 5.052B; 1%/9% are rounded readings — Δpp∈[7,9] maps to about 4.491B~5.774B; the cache-write-at-1h-$8 sensitivity 5.391B not adopted; the raw-total basis 72,166,562÷8%×4 weeks = 3.608B (weekly pool 902,082,025 raw) kept for comparison; factoryCredits 16,935,482 ≈ 212M credits/week; the two 4pp segments each carry 8.90M/9.14M tok per pp (2.7% apart) — effort only affects rate; Factory also has 5h and 30-day rolling windows — if the 30-day window is below 4× the weekly pool this value skews high; the Pro $20 / Plus $100 official pages state only about 1/10 and 1/5 of Max usage, not derived

stdout,证据链接:

Evidence URL
https://factory.ai/pricing https://factory.ai/pricing
droid-opus55-max-round1-2026-09-29.json /data/evidence/droid-opus55-max-round1-2026-09-29.json

证据的 /data/... 是快照保存的相对网站路径,CLI 原样保留该路径。

stdout,榜单与配置字段节选;完整输出另含 Board ID、Harness、Mode,以及每条配置的完整字段表。Benchmark source 与 Record note 优先使用配置中的 source_en 与 record_note_en:

Board ID Board Board snapshot
aa_intelligence_index Artificial Analysis Intelligence Index v4.3 2026-09-22
Configuration ID Variant Score Effort Estimated Self-reported Mapping confidence Quota effort matched
aa_intelligence_index:6df46e4e9119d1e1 Claude Opus 5.5 (max with fallback) 57.6224 Max false false Medium unverified
aa_intelligence_index:4c3fe13d36952d4f Claude Opus 5.5 (xhigh with fallback) 55.9874 xhigh false false Medium unverified
aa_intelligence_index:c4632b7e15588786 Claude Opus 5.5 (high with fallback) 53.5832 High false false Medium unverified
aa_intelligence_index:b44090b237d31264 Claude Opus 5.5 (medium with fallback) 51.2435 Medium false false Medium unverified
aa_intelligence_index:4dae1243cd80cd14 Claude Opus 5.5 (low with fallback) 42.3078 Low false false Medium unverified
Benchmark source: https://artificialanalysis.ai/leaderboards/models
Record note: v4.3.2 original precision (page shows integer 58). Released 9/22, all five tiers listed the same day and topped the board (next-highest Fable 5.1 max 53.35). A slug without a -max suffix is the max tier. The AA page notes releaseDate 2026-09-22; the other four tiers are marked 9-17 (per-tier measurement dates).
Mapping kind: model_configuration_reference
Mapping note: Exact served-model reference; quota-measurement effort is unverified.

stderr:

Snapshot: 2026-10-01 | Source: bundled
Board: aa_intelligence_index | Board snapshot: 2026-09-22
Total: 1 | Returned: 1 | Truncated: false

7. rap compare:比较指定价格点

用途:比较官方套餐与第三方渠道,不将同一模型视为同一套餐。保留输入次序,先展示两条核心记录,再逐项对齐采用依据;指定榜单时附加最高配置。至少需要两个不同的精确 Point ID。

rap compare 'claude_max_20x::claude-opus-5.5' \
  'droid_max::claude-opus-5.5' --board aa_intelligence_index

stdout,核心数据:

Model Company Channel Plan USD/MTok Monthly tokens Monthly fee USD Billing Confidence Point ID
Claude Opus 5.5 Anthropic Anthropic Claude Max 20x (9/14+) $0.00634156 31.538 B $200 Subscription Medium claude_max_20x::claude-opus-5.5
Claude Opus 5.5 Anthropic Factory Droid Max $0.0395883 5.052 B $200 Subscription Medium droid_max::claude-opus-5.5

stdout,比较字段节选:

Field Claude Max 20x (9/14+) Droid Max
Point ID claude_max_20x::claude-opus-5.5 droid_max::claude-opus-5.5
Monthly tokens, raw 31,538,000,000 5,052,000,000
USD/MTok, raw 0.0063415562 0.039588282
Original monthly fee 200 USD 200 USD
Workload Anthropic workload: 97% cache reads / 2.5% cache writes / 0.5% output Anthropic workload: 97% cache reads / 2.5% cache writes / 0.5% output
Data date 2026-09-22~2026-09-27 2026-09-29
Date kind sample sample
Date inherited from — —
Promotion until — —
Unmetered false false
Plan generation
Board aa_intelligence_index aa_intelligence_index
Board snapshot 2026-09-22 2026-09-22
Score 57.6224 57.6224
Effort Max Max
Variant Claude Opus 5.5 (max with fallback) Claude Opus 5.5 (max with fallback)
Configuration ID aa_intelligence_index:6df46e4e9119d1e1 aa_intelligence_index:6df46e4e9119d1e1
Estimated false false
Self-reported false false
Mapping confidence Medium Medium
Quota effort matched unverified unverified

stdout,Claude Max 的英文证据区域;与 Decision note 相同的 Note 不重复输出:

Point ID: claude_max_20x::claude-opus-5.5
Source:
X @MiaAI_lab tweet (screenshot provided by the maintainer): xHigh burned 1.0305B raw in 1h2m = ~75% of the 5h limit; claude-opus55-round1-2026-09-23.json; round10 item #15 Max 20x ≈5.5 full windows/week (claude-adoption-round10-2026-09-25.json); Claude Pro × Opus 5.5 adopted sample (issue #52 + Reddit segment) ×10; claude-opus55-max20x-round2-2026-09-30.json

Decision note:
30.17B→31.538B (2026-09-30 decision, Anthropic tier): each path converted to Max 20x weekly-quota percentage points, then merged per pp — (i) the MiaAI window worth $471.10 (write at 5m; closes against the tweet's $482.63) = 75% of a window ÷ 5.5 windows/week = 13.636 weekly pp, 32.981B alone; (ii) the Pro × Opus 5.5 merged worth $628.00 / 196.333 Pro pp ÷ Max 20x weekly quota = Pro×10 → 19.633 weekly pp, 30.536B alone; the two paths differ by 8%. Not adopted: Pro 5h/week 13.333% (7.5 windows) applied directly to Max 20x gives 44.97B (Max 20x windows 20×, week 10× — the window/week ratio differs from Pro's); the official-multiplier derivation with 3.75 windows/week gives 22.49B (the MiaAI window is only 15.5× a Pro window, not 20× — inconsistent with both multipliers at once); equal-weight mean 31.758B; the #65 mixed sample not yet handled. Original 301.7 decision text — new 30.17B: 5h pool 1.0305B÷~75% = 1.374B raw (the 9/22 release had already raised Pro/Max/Team 5h caps, so the window pool reflects release-period basis); implied weight 7.15/13.74 = 0.5204×Opus5 → monthly pool 157÷0.5204; the list-price blend ratio 0.5143 independently corroborates (alternative 30.528B differs by 1.2%); the sample closes at the new price $471.1≈tweet $482.63 (+2.4%); n=1 tweet with no panel, ~75% rounded reading (monthly interval about 28.3B~32.2B), effort only affects rate; if official weights deviate from the price ratio (Fable 6.5× precedent) it must be re-derived; direct weekly-pool panel measurement or post-reset controlled saturation could raise it to high
Evidence URL
claude-adoption-round10-2026-09-25.json /data/evidence/claude-adoption-round10-2026-09-25.json
claude-opus55-max20x-round2-2026-09-30.json /data/evidence/claude-opus55-max20x-round2-2026-09-30.json
claude-opus55-round1-2026-09-23.json /data/evidence/claude-opus55-round1-2026-09-23.json

其后输出 Point ID: droid_max::claude-opus-5.5 对应的 Source、Decision note 和 Evidence,内容与第 6 节相同;本页不重复这段长文本。比较保留原始采用理由,不自动归纳为推荐结论,也不合计同套餐不同模型的额度。

stderr:

Snapshot: 2026-10-01 | Source: bundled
Board: aa_intelligence_index | Board snapshot: 2026-09-22 | Config: best
Total: 2 | Returned: 2 | Truncated: false

8. rap info:快照与口径

用途:确认本次查询读哪份数据、数据规模与采用口径。这里的快照字段本身是命令数据,写 stdout。

rap info

stdout:

Field Value
Data source bundled
Data snapshot 2026-10-01
Dataset version 1
Price points 318
Models 74
Companies 18
Channels 17
Plans 95
Boards 8
Benchmark configurations 330
Price/configuration mappings 1854
Subscription price assumption Full use of the adopted allowance
Allowances across models in one plan Alternative use; do not add together
Generic month 4 weeks; vendor-specific monthly pools retain their adopted basis
Standard workload 97% cache reads / 2.5% fresh input / 0.5% output
Anthropic workload 97% cache reads / 2.5% cache writes / 0.5% output
Low-cache workload 85% cache reads / 14.5% fresh input / 0.5% output
Measured workload Raw sample tokens; not workload-normalized
Adopted exchange rate 1 USD = 6.7787 CNY
Exchange rate date 2026-09-04
Exchange rate source CFETS central parity, via Xinhua

stderr:

Total: 1 | Returned: 1 | Truncated: false

当前默认联网成功时,Data source 为 remote;ETag/304 复用或网络失败时读取有效缓存则为 cache: <绝对路径>,没有有效缓存时为 bundled。以下来源示意不代表固定快照日期:

Field Value
Data source remote
Data snapshot 2026-10-03

带 --data 时完全跳过联网和缓存,source 显示文件来源与解析后的路径;数据快照取文件中的 generatedAt,不取文件修改时间:

rap info --data ./web/public/data/site.json

在仓库根目录调用时,其余字段与上表一致,前两项为:

Field Value
Data source file: /workspace/real-api-pricing/web/public/data/site.json
Data snapshot 2026-10-01

9. 帮助与版本

rap --help

用途:发现入口及全局选项。帮助由 Commander 自动生成,不加载快照、不读写缓存、不联网。以下为当前英文帮助文本:

rap --help

stdout:

Usage: rap [options] [command]

Read model prices and monthly allowances with automatic offline fallback.

Options:
  -V, --version                     output the version number
  --data <site.json>                Read a specified local snapshot
  --format <format>                 Output format: table | json | csv (default:
                                    "table")
  -h, --help                        display help for command

Commands:
  price [options]                   Real price ranking, lowest first; includes
                                    metered APIs and unscored points.
  allowance [options]               Monthly allowance ranking, largest first;
                                    subscriptions with known monthly tokens.
  query [options]                   Query price points; optionally filter
                                    benchmark configurations.
  list                              Discover exact model, company, channel, plan
                                    and board IDs.
  show [options] <point-id>         Show price, allowance, evidence and all
                                    mapped benchmark configurations.
  compare [options] <point-ids...>  Compare at least two distinct price points,
                                    in input order.
  info                              Show snapshot, coverage and calculation
                                    conventions.

Examples:
  rap price --company Anthropic
  rap allowance --company Anthropic --fee-band 0-30
  rap list models --search claude
  rap query --board aa_intelligence_index --frontier --format json

Run rap <command> --help for command-specific options.

帮助与版本不会附加查询统计到 stderr。--format 面向数据命令,help/version 始终输出文本。

rap price --help

rap price --help

stdout:

Usage: rap price [options]

Real price ranking, lowest first; includes metered APIs and unscored points.

Options:
  --company <id>          Exact model developer ID; repeatable
  --vendor <id>           Alias of --company; values are combined
  --channel <name>        Exact access channel; repeatable
  --model <id>            Exact model ID; repeatable
  --plan <id>             Exact plan ID; repeatable
  --billing <kind>        subscription | metered; repeatable
  --confidence <level>    high | medium | low; repeatable
  --search <text>         Whole-substring website search (default: "")
  --min-fee <USD>         Inclusive lower monthly-fee bound
  --max-fee <USD>         Inclusive upper monthly-fee bound
  --min-price <USD/MTok>  Inclusive lower real-price bound
  --max-price <USD/MTok>  Inclusive upper real-price bound
  --min-tokens <integer>  Inclusive lower monthly-token bound
  --max-tokens <integer>  Inclusive upper monthly-token bound
  --board <id>            Independent benchmark board
  --harness <name>        Benchmark harness; requires --board; repeatable
  --effort <name>         Reasoning effort; requires --board; repeatable
  --mode <name>           Service mode; requires --board; repeatable
  --limit <integer>       Return at most this many rows; default: all
  -h, --help              display help for command

Global Options:
  -V, --version           output the version number
  --data <site.json>      Read a specified local snapshot
  --format <format>       Output format: table | json | csv (default: "table")

Global options: --data <site.json> --format table|json|csv --version
Use query --view table for custom sorting, score filters and configuration expansion.

rap allowance --help 同样列出公共筛选,增加 --fee-band all|0-30|30-100|100-300,说明只保留有月额度的订阅且额度降序。rap query --help 列出 --view、--sort 和评测选项的依赖;rap list <resource> --help 按第 4 节的适用矩阵列参数;show/compare 的查询选项只有 --board,不接受排名的过滤、排序与 limit。

rap --version

当前软件版本为 0.1.0;内置数据日期另由 rap info 查询。

rap --version

stdout:

rap 0.1.0

10. 空结果与参数错误

查询无匹配记录是成功,stdout 显示空结果,退出码为 0:

rap price --company Anthropic --search no-such-model-123

stdout:

No matching results.

stderr:

Snapshot: 2026-10-01 | Source: bundled | View: price
Total: 0 | Returned: 0 | Truncated: false
Order: USD/MTok ascending; ties by Point ID.

精确 ID、参数枚举或依赖错误的退出码为 2;stdout 为空,错误只写 stderr。例如:

rap price --company Factory
error: Unknown company ID "Factory". Factory is a Channel. Use --channel Factory.
rap price --company Anthropic --sort fee:asc
error: unknown option '--sort'
(Did you mean one of --effort, --format?)

Commander 随后附加 price 的帮助。自定义排序使用 rap query --view table --sort fee:asc。

rap query --frontier
error: --frontier requires --board.
rap list boards --company Anthropic
error: unknown option '--company'

Commander 随后附加 list boards 的帮助。

显式 --data 文件或最终内置兜底数据读取、格式错误退出码为 1,不能输出半份成功结果;默认网络失败、超时或线上数据无效会自动回退,不作为查询错误输出。JSON 和 CSV 使用与表格相同的记录集合,空 JSON 返回 rows: [],空 CSV 保留表头;诊断仍写 stderr。CLI CSV 保持对应命令排序;网站排名视图的“Table · CSV”采用明细表排序(allowance 下载也按价格),固定列 schema 也不同,因此按 Point ID 对账共享原始字段,不要求文件逐字一致。比较榜单引用字段时显式选择同一榜单。完整导出契约见 CLI 设计规范。