@angelyeye/dsh-cost-tracker 1.8.5 → 1.8.6
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +31 -0
- package/README.en.md +12 -2
- package/README.md +12 -2
- package/index.js +32 -16
- package/package.json +1 -1
- package/pricing.js +63 -13
package/CHANGELOG.md
CHANGED
|
@@ -2,6 +2,37 @@
|
|
|
2
2
|
|
|
3
3
|
本文件用中文记录 dsh-cost-tracker 的版本变更。
|
|
4
4
|
|
|
5
|
+
## v1.8.6(2026-09-15)
|
|
6
|
+
|
|
7
|
+
**修复:计费规则与 DeepSeek 官方定价的两处口径偏差(金额口径 + 标记口径)**
|
|
8
|
+
|
|
9
|
+
核查基准:官方价格卡 <https://api-docs.deepseek.com/zh-cn/quick_start/pricing> 与发布通告 <https://api-docs.deepseek.com/zh-cn/news/news260910>。核查方法:用独立脚本按官方牌价重算全部已入库记录,与插件记账逐条比对(1985 条记录金额差额 `0.000000`,说明**数值本来就是对的一分不差**;以下两处修复的是口径与标记)。
|
|
10
|
+
|
|
11
|
+
- **V4-Pro 路由生效时刻提前了 4 天(金额口径,会低估花费)**
|
|
12
|
+
- 旧实现把 `deepseek-v4-pro → V4.1 Flash` 的路由与 Flash 调价合并成同一时刻(北京时间 2026-09-10 12:00),但官方通告的措辞是:「北京时间 **2026 年 9 月 14 日 12:00** 之后……用户访问 `deepseek-v4-pro` 的请求将全部路由到 V4.1 Flash,并按 V4.1 Flash 单价计费」(价格卡脚注 (2) 同口径)。
|
|
13
|
+
- 影响:2026-09-10 12:00 ~ 09-14 12:00 之间的 V4-Pro 调用会被**低估约 4.5 倍**(9/27/0.30 → 2/8/0.04)。本机记录该窗口内无 V4-Pro 调用,故无历史金额损失,但属潜伏缺陷。
|
|
14
|
+
- 修法:新增 `V41_PRO_ROUTE_AT`(`2026-09-14T04:00:00Z`)与时代 `v41pro`,把路由从 `v41` 拆出来;`v41` 时代保留 V4-Pro 自有牌价(9/27/0.30),避免这 4 天落入 provider 兜底而被误标「估算」。
|
|
15
|
+
|
|
16
|
+
- **官方现役模型名 `deepseek-flash` 不在精确单价表内(标记口径,已在真实数据中触发)**
|
|
17
|
+
- 官方脚注 (1):「模型名请使用 `deepseek-flash`」。旧实现的规范名是 `deepseek-v4.1-flash`(官方文档中并不可调用的写法),于是宿主实际上报的 `deepseek-flash` 落入 provider 兜底分支——**金额恰好相同,但被标记为「估算」**。本机 1402 条 `deepseek-flash` 记录 100% 命中该问题,云端 `cost_basis` 也随之为 `estimated`,污染「估算占比 / 口径漂移」统计。
|
|
18
|
+
- 修法:`V41_FLASH_MODEL` 改为 `deepseek-flash`(官方现役名),并新增 `MODEL_ALIASES` 别名归一表,使 `deepseek-v4.1-flash` / `deepseek-v4-1-flash` / `deepseek-v41-flash` / `deepseek_flash` 等等价写法解析到同一规范名(单价表本身不重复列项)。路由目标与入账名统一为 `deepseek-flash`。
|
|
19
|
+
|
|
20
|
+
**修复:`cost_recompute` 不再漏判「仅标记变化」的记录**
|
|
21
|
+
- 变更判定此前只看费用 / 模型名 / 档位,于是「模型名从兜底升为精确档、金额分毫不变」的记录不会被订正。现纳入 `estimated` / `subscription` 标记比对,返回值新增 `estimatedFlips` 计数,工具输出会明确写出「其中 N 条仅订正『估算』标记,金额不变」,避免补账看起来「改了 0 条」。
|
|
22
|
+
|
|
23
|
+
**变更**
|
|
24
|
+
- `export const V41_FLASH_MODEL = 'deepseek-flash'`(原 `'deepseek-v4.1-flash'`);新增导出 `V41_PRO_ROUTE_AT` / `MODEL_ALIASES`。
|
|
25
|
+
- `PRICE_ERAS` 由 2 版增至 3 版:`legacy` / `v41`(09-10 12:00 起,含 V4-Pro 自有牌价)/ `v41pro`(09-14 12:00 起,含 V4-Pro 路由);`v41` 时代新增 `proRouteSince` 说明字段。
|
|
26
|
+
- `cost_prices` 的 `eras` / 渲染文本同步展示 V4-Pro 路由时刻与官方现役名口径。
|
|
27
|
+
|
|
28
|
+
**测试**
|
|
29
|
+
- `test/pricing.test.js`:新增 V4-Pro 路由时刻边界用例(`12:00:00.000` 前一毫秒仍按自有牌价、整点起路由)、`deepseek-flash` 精确命中与别名归一用例。
|
|
30
|
+
- `test/recompute.test.js`:补账样本改为覆盖「路由前窗口金额不变」「路由后改写为 `deepseek-flash`」「误标估算记录被订正」三类,新增 `estimatedFlips` 断言。
|
|
31
|
+
|
|
32
|
+
**计费影响**
|
|
33
|
+
- **已入库记录金额零变化**(数值本即正确);仅 `estimated` 标记、V4-Pro 在 9/10–9/14 窗口的口径、以及历史 `deepseek-v4.1-flash` 记录(本机 22 条)与 `deepseek-flash` 的桶名归并发生变化。
|
|
34
|
+
- 涉及历史桶名/标记的机器建议执行一次 `cost_recompute`(默认试算,`apply: true` 落盘),幂等可重复执行。
|
|
35
|
+
|
|
5
36
|
## v1.8.5(2026-09-15)
|
|
6
37
|
|
|
7
38
|
**修复**
|
package/README.en.md
CHANGED
|
@@ -29,10 +29,11 @@
|
|
|
29
29
|
| | Feature | Description |
|
|
30
30
|
| --- | --- | --- |
|
|
31
31
|
| 💰 | **Cost tracking** | Every API call is recorded automatically: input / output / cache-hit / cache-write tokens and cost (cache write billed at the cache-hit price), aggregated by day and by model |
|
|
32
|
-
| ⏰ | **Peak/off-peak pricing** | Built-in price table; peak windows (Mon–Fri 9:00–12:00, 14:00–18:00 Beijing time) vs half-price off-peak handled automatically; **weekends are fully off-peak**; local models (e.g. ollama) count as 0; the table is **versioned by pricing era** and switches automatically to V4.1 Flash rates at 2026-09-10 12:00 Beijing time (peak 2.0 / 0.04 / 8.0), routing
|
|
32
|
+
| ⏰ | **Peak/off-peak pricing** | Built-in price table; peak windows (Mon–Fri 9:00–12:00, 14:00–18:00 Beijing time) vs half-price off-peak handled automatically; **weekends are fully off-peak**; local models (e.g. ollama) count as 0; the table is **versioned by pricing era** and switches automatically to V4.1 Flash rates at 2026-09-10 12:00 Beijing time (peak 2.0 / 0.04 / 8.0), routing the legacy V4-Flash family to V4.1 Flash billing per the official rules; **V4-Pro is routed from 2026-09-14 12:00 only** (before that it keeps its own rates 9.0 / 27.0 / 0.30) |
|
|
33
33
|
| 🔔 | **Peak-price notice** | A "Peak/off-peak pricing & notice" panel in Settings: current tier + countdown strip, style switch (**compact single-row · 24h-proportional** / **ring dial · phase dot**), an optional **two-row compact** stacked layout for the compact style (bar-above-text / text-above-bar), a "Show time" toggle, popup alert + browser notification before a tier switch, lead time, popup position (bottom-right/center), alert type; a persistent strip in the sidebar footer (adapts to rail/collapsed). Mirrors `dsh-cost-meter` |
|
|
34
34
|
| 📌 | **Six overview cards** | Six cards at the top of Settings: Today / This month / Total spend / API requests / Tokens / **Account balance**. The three spend cards **exclude subscription equivalent cost** (shown as an annotation instead) |
|
|
35
|
-
| 👁️ | **Vision model** | Supports `deepseek-v4-flash-vision-exp`: priced as flash in the legacy era, billed at V4.1 Flash rates from 2026-09-10 12:00; images are converted to tokens per the official rule (≤384 tokens each, billed per API usage) |
|
|
35
|
+
| 👁️ | **Vision model** | Supports `deepseek-v4-flash-vision-exp`: priced as flash in the legacy era, billed at V4.1 Flash rates from 2026-09-10 12:00 (that legacy id is retired; requests are served by V4.1 Flash); images are converted to tokens per the official rule (≤384 tokens each, billed per API usage) |
|
|
36
|
+
| 🏷️ | **Official model-name alignment** | Billing recognises the official current name **`deepseek-flash`** (price-card note (1): "use the model name `deepseek-flash`"); older spellings such as `deepseek-v4.1-flash` normalize to the same rate, so an official rename can never silently drop usage into the fallback estimate |
|
|
36
37
|
| 📊 | **Visual dashboard** | A new "Cost Statistics" page in Settings: overview cards, cost bar charts (by peak period / by model), per-model request & token charts — **all with hover tooltips** |
|
|
37
38
|
| 📈 | **Subscription quota** | Kimi Coding Plan and similar subscriptions: weekly quota, 5-hour rolling window limit, pay-as-you-go-equivalent cost for reference |
|
|
38
39
|
| 💳 | **Balance lookup** | One-click DeepSeek account balance (total / topped-up / granted / status) |
|
|
@@ -236,6 +237,15 @@ Example: `curl -X POST http://127.0.0.1:3080/api/cost-tracker/summary -d '{}'`
|
|
|
236
237
|
|
|
237
238
|
## Changelog
|
|
238
239
|
|
|
240
|
+
### v1.8.6 (2026-09-15)
|
|
241
|
+
|
|
242
|
+
**Fixed: two basis mismatches against DeepSeek's official pricing** (checked against the official [price card](https://api-docs.deepseek.com/zh-cn/quick_start/pricing) and the [V4.1 Flash announcement](https://api-docs.deepseek.com/zh-cn/news/news260910))
|
|
243
|
+
|
|
244
|
+
- **The V4-Pro routing date was 4 days early (understated spend)**: the old code merged `deepseek-v4-pro → V4.1 Flash` routing with the Flash price change (09-10 12:00), but the official wording is "after **2026-09-14 12:00 Beijing time**". V4-Pro calls in that 4-day window were understated by roughly 4.5×. The eras are now split with `v41pro` (`V41_PRO_ROUTE_AT = 2026-09-14T04:00:00Z`), and era `v41` keeps V4-Pro's own rates (9 / 27 / 0.30).
|
|
245
|
+
- **The official current model name `deepseek-flash` was missing from the exact table (records mislabelled "estimated")**: the official docs say "use the model name `deepseek-flash`", while the code used the non-callable `deepseek-v4.1-flash` as its canonical name — so the real name fell through to the provider fallback. The amount was exactly right, but every record was flagged `estimated: true` (1402 local records affected). `deepseek-flash` is now canonical, with an alias table (`MODEL_ALIASES`) mapping equivalent spellings.
|
|
246
|
+
- **`cost_recompute` no longer skips "flag-only" changes**: it now reports `estimatedFlips` and states explicitly that N records only had their "estimated" flag corrected with no amount change.
|
|
247
|
+
- **Stored amounts do not change**: only the `estimated` flag, the V4-Pro basis inside the 09-10…09-14 window, and the bucket merge of historical `deepseek-v4.1-flash` records (22 locally) into `deepseek-flash`. Running `cost_recompute` once is recommended (dry-run by default, `apply: true` to persist).
|
|
248
|
+
|
|
239
249
|
### v1.6.0 (2026-09-10)
|
|
240
250
|
|
|
241
251
|
**Added**
|
package/README.md
CHANGED
|
@@ -29,10 +29,11 @@
|
|
|
29
29
|
| | 功能 | 说明 |
|
|
30
30
|
| --- | --- | --- |
|
|
31
31
|
| 💰 | **花费统计** | 每一次 API 调用自动记账:输入 / 输出 / 缓存命中 / 缓存写入 Tokens 与费用(缓存写入按缓存命中价计),按天、按模型聚合 |
|
|
32
|
-
| ⏰ | **峰谷定价** | 内置单价表,高峰时段(北京时间周一至周五 9:00–12:00、14:00–18:00)与闲时半价自动区分,**周末全天计为闲时**,本地模型(ollama 等)计 0;**单价表按「计费时代」分版**,北京时间 2026-09-10 12:00 起自动切换为 V4.1 Flash 新价(2.0 / 0.04 / 8.0 高峰)
|
|
32
|
+
| ⏰ | **峰谷定价** | 内置单价表,高峰时段(北京时间周一至周五 9:00–12:00、14:00–18:00)与闲时半价自动区分,**周末全天计为闲时**,本地模型(ollama 等)计 0;**单价表按「计费时代」分版**,北京时间 2026-09-10 12:00 起自动切换为 V4.1 Flash 新价(2.0 / 0.04 / 8.0 高峰),并按官方规则把旧 V4-Flash 系请求路由到 V4.1 Flash 计费;**V4-Pro 自 2026-09-14 12:00 起才路由**(此前仍按自有牌价 9.0 / 27.0 / 0.30) |
|
|
33
33
|
| 🔔 | **峰谷计价提示** | 设置页「峰谷计价与提示」面板:当前档位/距下次切换倒计时时段条、样式切换(**简洁单行·按24h比例** / **环形表盘·相位色点**)、简洁样式可选**双行紧凑**(上下布局,条上文下/文上条下)、「显示时间」开关、峰/谷切换前弹窗提醒与浏览器系统通知、提前提醒分钟、弹窗位置(右下角/屏幕中心)、提醒类型;侧边栏底部常驻显示时段条(窄栏/展开自适应)。对齐 `dsh-cost-meter` 交互 |
|
|
34
34
|
| 📌 | **六组概览卡** | 设置页顶部六张卡:今日费用 / 本月费用 / 总花费 / API 请求次数 / Tokens / **总余额**。前三个金额卡**不含订阅会员等效费用**,订阅以附注展示 |
|
|
35
|
-
| 👁️ | **视觉模型** | 支持 `deepseek-v4-flash-vision-exp`:legacy 时代单价与 flash 一致,2026-09-10 12:00 起随 V4.1 Flash
|
|
35
|
+
| 👁️ | **视觉模型** | 支持 `deepseek-v4-flash-vision-exp`:legacy 时代单价与 flash 一致,2026-09-10 12:00 起随 V4.1 Flash 新价计费(该旧名已下线,请求由 V4.1 Flash 提供);图片按官方规则换算 token(每张上限 384 个,以接口用量计费) |
|
|
36
|
+
| 🏷️ | **官方模型名对齐** | 计费支持**官方现役名 `deepseek-flash`**(价格卡脚注 (1):「模型名请使用 `deepseek-flash`」),旧写法 `deepseek-v4.1-flash` 等归一化后命中同一档价,避免因官方改名而落入兜底估算 |
|
|
36
37
|
| 📊 | **可视化仪表盘** | 设置页新增「花费统计」:概览卡片、消费柱状图(按峰谷/按模型)、分模型的请求次数与 Tokens 图表,**全部支持鼠标悬停查看明细** |
|
|
37
38
|
| 🔥 | **用量热力图** | 设置页新增「Token 用量统计」:类 Codex 的 **26 周日用量方格热图**,按天着色(输入 / 缓存 / 输出 / 费用),悬停看当日明细、今天高亮描边,顶部显示全时段累计 |
|
|
38
39
|
| 📈 | **订阅配额监控** | Kimi Coding Plan 等订阅套餐:本周配额、5 小时滚动窗口限额、等效按量费用参考 |
|
|
@@ -237,6 +238,15 @@ POST /api/cost-tracker/export 导出 CSV
|
|
|
237
238
|
|
|
238
239
|
## 更新记录
|
|
239
240
|
|
|
241
|
+
### v1.8.6(2026-09-15)
|
|
242
|
+
|
|
243
|
+
**修复:计费规则与官方定价的两处口径偏差**(核查基准:官方[价格卡](https://api-docs.deepseek.com/zh-cn/quick_start/pricing) + [V4.1 Flash 发布通告](https://api-docs.deepseek.com/zh-cn/news/news260910))
|
|
244
|
+
|
|
245
|
+
- **V4-Pro 路由时刻提前了 4 天(会低估花费)**:旧实现把 `deepseek-v4-pro → V4.1 Flash` 的路由与 Flash 调价合并为同一时刻(09-10 12:00),而官方口径是「北京时间 **2026-09-14 12:00** 之后」才路由。这 4 天内的 V4-Pro 调用会被低估约 4.5 倍。现拆出时代 `v41pro`(`V41_PRO_ROUTE_AT = 2026-09-14T04:00:00Z`),`v41` 时代保留 V4-Pro 自有牌价 9/27/0.30。
|
|
246
|
+
- **官方现役名 `deepseek-flash` 未进精确表(会被误标「估算」)**:官方要求使用 `deepseek-flash`,旧代码的规范名却是不可调用的 `deepseek-v4.1-flash`,导致现役名走 provider 兜底——金额分毫不差,但记录被标 `estimated: true`(本机 1402 条全部中招)。现以 `deepseek-flash` 为规范名,并加别名归一表(`MODEL_ALIASES`)。
|
|
247
|
+
- **`cost_recompute` 不再漏判「仅标记变化」**:返回值新增 `estimatedFlips`,输出会写明「其中 N 条仅订正『估算』标记,金额不变」。
|
|
248
|
+
- **已入库记录金额零变化**:仅 `estimated` 标记、V4-Pro 在 09-10~09-14 的口径、以及历史 `deepseek-v4.1-flash` 记录(22 条)与 `deepseek-flash` 的桶名归并变化;建议执行一次 `cost_recompute`(默认试算,`apply: true` 落盘)。
|
|
249
|
+
|
|
240
250
|
### v1.6.0(2026-09-10)
|
|
241
251
|
|
|
242
252
|
**新增**
|
package/index.js
CHANGED
|
@@ -10,7 +10,7 @@ import { readFileSync, writeFileSync, mkdirSync, renameSync, existsSync, realpat
|
|
|
10
10
|
import { join, dirname } from 'node:path'
|
|
11
11
|
import { homedir } from 'node:os'
|
|
12
12
|
import { createStore, collectTotals, DETAIL_DAYS, MAX_AXIS_DAYS } from './store.js'
|
|
13
|
-
import { PRICE_ERAS, V41_EFFECTIVE_AT, exactModelsAt, eraAt, SUBSCRIPTION_RATES, PROVIDER_RATES, GENERIC_RATES, PEAK_WINDOWS, PEAK_HOUR_WINDOWS, isPeak, peakPhaseAt, priceFor, computeCost, normalizeTokens } from './pricing.js'
|
|
13
|
+
import { PRICE_ERAS, V41_EFFECTIVE_AT, V41_PRO_ROUTE_AT, exactModelsAt, eraAt, SUBSCRIPTION_RATES, PROVIDER_RATES, GENERIC_RATES, PEAK_WINDOWS, PEAK_HOUR_WINDOWS, isPeak, peakPhaseAt, priceFor, computeCost, normalizeTokens } from './pricing.js'
|
|
14
14
|
import {
|
|
15
15
|
normalizePeakConfig, defaultPeakConfig, peakEffective,
|
|
16
16
|
normalizeCloudConfig, defaultCloudConfig, normalizePluginConfig,
|
|
@@ -19,7 +19,7 @@ import { createSyncEngine, setPluginVersion, SOURCE as SYNC_SOURCE, SYNC_VERSION
|
|
|
19
19
|
import { Schema } from './schema.js'
|
|
20
20
|
|
|
21
21
|
/** 插件版本(写入上报信封,便于云端排查版本差异) */
|
|
22
|
-
const PLUGIN_VERSION = '1.8.
|
|
22
|
+
const PLUGIN_VERSION = '1.8.6'
|
|
23
23
|
setPluginVersion(PLUGIN_VERSION)
|
|
24
24
|
|
|
25
25
|
/** 「设置 → 插件 → 插件配置」里的卡片字段(与 settings 命名空间一致) */
|
|
@@ -105,8 +105,9 @@ export default {
|
|
|
105
105
|
// ---------- price tables (CNY per 1M tokens) ----------
|
|
106
106
|
// 单价表 / 峰谷 / 费用计算集中在 pricing.js(纯模块,可独立测试);
|
|
107
107
|
// 单价按「计费时代」分版(PRICE_ERAS):按记录时间戳选版,故历史记录口径不变。
|
|
108
|
-
// V4.1 Flash 价(北京时间 2026-09-10 12:00
|
|
109
|
-
//
|
|
108
|
+
// V4.1 Flash 价(北京时间 2026-09-10 12:00 起)生效后,旧 V4-Flash 系的请求路由到
|
|
109
|
+
// V4.1 Flash 计费;V4-Pro 自北京时间 2026-09-14 12:00 起才路由(官方通告口径)。
|
|
110
|
+
// 记录一律以官方现役模型名 `deepseek-flash` 入账。
|
|
110
111
|
// 视觉模型 deepseek-v4-flash-vision-exp 的图片 token 由接口 usage 计入 inputTokens
|
|
111
112
|
|
|
112
113
|
// ---------- state ----------
|
|
@@ -286,7 +287,7 @@ export default {
|
|
|
286
287
|
const model = toStr(options && options.model)
|
|
287
288
|
if (!provider && !model) return
|
|
288
289
|
const np = normProvider(provider)
|
|
289
|
-
// 按「调用发生的时刻」选单价版本(跨 2026-09-10 12:00 自动切换,无需重启)。
|
|
290
|
+
// 按「调用发生的时刻」选单价版本(跨 2026-09-10 12:00 / 2026-09-14 12:00 自动切换,无需重启)。
|
|
290
291
|
const price = priceFor(np, model, ts)
|
|
291
292
|
// 峰谷计费开关:随配置峰谷启用 + 生效时间门控;未启用时按非峰谷档(平价)计费。
|
|
292
293
|
const peak = peakEffective(peakConfig, ts) ? isPeak(ts) : false
|
|
@@ -294,8 +295,8 @@ export default {
|
|
|
294
295
|
// prompt_tokens 中(每张≤384 tokens),由 normalizeTokens 归入 input
|
|
295
296
|
const tokens = normalizeTokens(usage)
|
|
296
297
|
store.add({
|
|
297
|
-
//
|
|
298
|
-
//
|
|
298
|
+
// 被路由的请求以「实际计费模型规范名」入账(如 V4-Pro → deepseek-flash、
|
|
299
|
+
// 旧名 deepseek-v4-flash → deepseek-flash),使按模型聚合看到的就是真实计费口径。
|
|
299
300
|
ts, provider, model: price.model || model,
|
|
300
301
|
sessionId: toStr(options && options.sessionId),
|
|
301
302
|
purpose: toStr(options && options.purpose),
|
|
@@ -536,6 +537,7 @@ export default {
|
|
|
536
537
|
// 全部价格时代(含生效时刻与路由规则),供工具/接口展示
|
|
537
538
|
eras: PRICE_ERAS.map((e) => ({ id: e.id, label: e.label, since: e.since, models: e.models, routes: e.routes || {} })),
|
|
538
539
|
v41EffectiveAt: V41_EFFECTIVE_AT,
|
|
540
|
+
v41ProRouteAt: V41_PRO_ROUTE_AT,
|
|
539
541
|
subscription: SUBSCRIPTION_RATES,
|
|
540
542
|
providers: PROVIDER_RATES,
|
|
541
543
|
generic: GENERIC_RATES,
|
|
@@ -780,7 +782,7 @@ export default {
|
|
|
780
782
|
else if (Number.isFinite(a.since) && a.since > 0) since = a.since
|
|
781
783
|
const apply = a.apply === true
|
|
782
784
|
const byModel = {}
|
|
783
|
-
let scanned = 0, changed = 0, oldCost = 0, newCost = 0
|
|
785
|
+
let scanned = 0, changed = 0, oldCost = 0, newCost = 0, estimatedFlips = 0
|
|
784
786
|
for (const r of records) {
|
|
785
787
|
if (!(r.ts >= since)) continue
|
|
786
788
|
scanned += 1
|
|
@@ -790,27 +792,34 @@ export default {
|
|
|
790
792
|
const cost = computeCost(price.rates, price.tiered, peak, r.tokens)
|
|
791
793
|
const model = price.model || r.model
|
|
792
794
|
const period = price.tiered ? (peak ? 'peak' : 'off-peak') : 'flat'
|
|
795
|
+
const estimated = price.estimated === true
|
|
796
|
+
const subscription = price.subscription === true
|
|
793
797
|
oldCost += r.cost
|
|
794
798
|
newCost += cost
|
|
795
|
-
|
|
799
|
+
// 变更判定含 estimated/subscription:模型名从兜底口径升为精确档时,
|
|
800
|
+
// 金额可能分毫不变,但「估算」标记必须一并订正(否则看板长期误报估算占比)。
|
|
801
|
+
const flip = (r.estimated === true) !== estimated || (r.subscription === true) !== subscription
|
|
802
|
+
if (!(Math.abs(cost - r.cost) > 1e-9 || model !== r.model || period !== r.period || flip)) continue
|
|
796
803
|
changed += 1
|
|
804
|
+
if (flip) estimatedFlips += 1
|
|
797
805
|
const key = r.provider + '|' + r.model + '|' + model
|
|
798
|
-
const m = byModel[key] || (byModel[key] = { provider: r.provider, from: r.model, to: model, calls: 0, oldCost: 0, newCost: 0 })
|
|
806
|
+
const m = byModel[key] || (byModel[key] = { provider: r.provider, from: r.model, to: model, calls: 0, oldCost: 0, newCost: 0, flips: 0 })
|
|
799
807
|
m.calls += 1
|
|
800
808
|
m.oldCost += r.cost
|
|
801
809
|
m.newCost += cost
|
|
810
|
+
if (flip) m.flips += 1
|
|
802
811
|
if (apply) {
|
|
803
812
|
r.cost = cost
|
|
804
813
|
r.model = model
|
|
805
814
|
r.period = period
|
|
806
|
-
r.estimated =
|
|
807
|
-
r.subscription =
|
|
815
|
+
r.estimated = estimated
|
|
816
|
+
r.subscription = subscription
|
|
808
817
|
}
|
|
809
818
|
}
|
|
810
819
|
if (apply && changed > 0) persistNow()
|
|
811
820
|
const rows = Object.keys(byModel).map(k => {
|
|
812
821
|
const m = byModel[k]
|
|
813
|
-
return { provider: m.provider, from: m.from, to: m.to, calls: m.calls, oldCost: r4(m.oldCost), newCost: r4(m.newCost), delta: r4(m.newCost - m.oldCost) }
|
|
822
|
+
return { provider: m.provider, from: m.from, to: m.to, calls: m.calls, oldCost: r4(m.oldCost), newCost: r4(m.newCost), delta: r4(m.newCost - m.oldCost), estimatedFlips: m.flips }
|
|
814
823
|
}).sort((x, y) => Math.abs(y.delta) - Math.abs(x.delta))
|
|
815
824
|
return {
|
|
816
825
|
ok: true,
|
|
@@ -819,11 +828,16 @@ export default {
|
|
|
819
828
|
era: eraAt(since).id,
|
|
820
829
|
scanned,
|
|
821
830
|
changed,
|
|
831
|
+
estimatedFlips,
|
|
822
832
|
oldCost: r4(oldCost),
|
|
823
833
|
newCost: r4(newCost),
|
|
824
834
|
delta: r4(newCost - oldCost),
|
|
825
835
|
byModel: rows,
|
|
826
|
-
note: changed === 0
|
|
836
|
+
note: changed === 0
|
|
837
|
+
? '没有需要重算的记录'
|
|
838
|
+
: (apply
|
|
839
|
+
? '已重算并落盘' + (estimatedFlips ? '(其中 ' + estimatedFlips + ' 条仅订正「估算」标记,金额不变)' : '')
|
|
840
|
+
: '试算结果,未落盘(传 apply: true 生效)'),
|
|
827
841
|
}
|
|
828
842
|
}
|
|
829
843
|
|
|
@@ -1115,6 +1129,8 @@ export default {
|
|
|
1115
1129
|
}
|
|
1116
1130
|
}
|
|
1117
1131
|
lines.push('', '当前生效:' + v.eraLabel + '(era=' + v.era + ')')
|
|
1132
|
+
lines.push('模型名口径:官方现役名为 deepseek-flash(被路由的请求一律以此名入账);deepseek-v4.1-flash 等写法归一化后命中同一档。')
|
|
1133
|
+
lines.push('V4-Pro 路由生效:' + when(v.v41ProRouteAt) + '(此前 deepseek-v4-pro 仍按 V4-Pro 自有牌价计费)。')
|
|
1118
1134
|
lines.push('视觉模型 deepseek-v4-flash-vision-exp:图片按官方规则换算 token(每张上限 384),以接口用量计费(已含在 inputTokens 内)。')
|
|
1119
1135
|
lines.push('kimi-coding(订阅等效,估算):输入 6.5 / 缓存命中(含缓存写入)1.1 / 输出 27.0')
|
|
1120
1136
|
lines.push('缓存写入(cache write)按缓存命中价计费,与官方规则一致。其他 provider 兜底为估算平价(openai 10/30/5,anthropic 15/75/1.5,gemini 2.5/10/0.625,未知 2/8/0.5);ollama/local 为 0。')
|
|
@@ -1156,11 +1172,11 @@ export default {
|
|
|
1156
1172
|
const when = new Date(v.since + 28800000).toISOString().replace('T', ' ').slice(0, 16).replace(/-/g, '/')
|
|
1157
1173
|
const lines = [
|
|
1158
1174
|
'费用重算(自 ' + when + ' 北京起 · era=' + v.era + ')',
|
|
1159
|
-
'扫描 ' + v.scanned + ' 条,需修正 ' + v.changed + ' 条',
|
|
1175
|
+
'扫描 ' + v.scanned + ' 条,需修正 ' + v.changed + ' 条' + (v.estimatedFlips ? '(其中 ' + v.estimatedFlips + ' 条仅订正「估算」标记,金额不变)' : ''),
|
|
1160
1176
|
'合计:¥' + v.oldCost + ' → ¥' + v.newCost + '(' + (v.delta >= 0 ? '+' : '') + v.delta + ')',
|
|
1161
1177
|
]
|
|
1162
1178
|
for (const m of (v.byModel || []).slice(0, 8)) {
|
|
1163
|
-
lines.push(' ' + m.from + (m.to !== m.from ? ' → ' + m.to : '') + ':' + m.calls + ' 次 · ¥' + m.oldCost + ' → ¥' + m.newCost)
|
|
1179
|
+
lines.push(' ' + m.from + (m.to !== m.from ? ' → ' + m.to : '') + ':' + m.calls + ' 次 · ¥' + m.oldCost + ' → ¥' + m.newCost + (m.estimatedFlips ? ' · 标记订正 ' + m.estimatedFlips + ' 条' : ''))
|
|
1164
1180
|
}
|
|
1165
1181
|
lines.push(v.note)
|
|
1166
1182
|
return [{ type: 'text', text: lines.join('\n') }]
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@angelyeye/dsh-cost-tracker",
|
|
3
|
-
"version": "1.8.
|
|
3
|
+
"version": "1.8.6",
|
|
4
4
|
"description": "DeepSeek Harness (DSH) plugin: LLM token usage & cost tracking with a settings dashboard, agent tools, HTTP API, persistent storage and multi-device cloud sync.",
|
|
5
5
|
"type": "module",
|
|
6
6
|
"main": "index.js",
|
package/pricing.js
CHANGED
|
@@ -7,10 +7,16 @@
|
|
|
7
7
|
// 空闲时段 = 高峰 × 0.5。周末(周六/周日)全天计入空闲时段(该窗口两版价通用)。
|
|
8
8
|
// - **单价随时间调整**:故单价表按生效时刻分版(见 PRICE_ERAS),按记录时间戳选版。
|
|
9
9
|
// · legacy :V4-Flash 3.0/9.0/0.10、V4-Pro 9.0/27.0/0.30
|
|
10
|
-
// · v41 :V4.1 Flash 2.0/8.0/0.04(北京时间 2026-09-10 12:00
|
|
11
|
-
//
|
|
12
|
-
//
|
|
10
|
+
// · v41 :V4.1 Flash 2.0/8.0/0.04(北京时间 2026-09-10 12:00 起生效)。
|
|
11
|
+
// 官方现役模型名为 `deepseek-flash`;旧名 `deepseek-v4-flash` 与
|
|
12
|
+
// `deepseek-v4-flash-vision-exp` 已下线但仍可调用,请求由 V4.1 Flash
|
|
13
|
+
// 提供服务并按 Flash 价计费,故一并路由到 `deepseek-flash`。
|
|
14
|
+
// · v41pro :同一档单价,**V4-Pro 路由在此才开始**:官方通告为「北京时间 2026-09-14
|
|
15
|
+
// 12:00 之后,deepseek-v4-pro 的请求全部路由到 V4.1 Flash 并按 V4.1 Flash
|
|
16
|
+
// 单价计费」(价格卡脚注 (2) 同口径)。9/10 12:00 ~ 9/14 12:00 之间的
|
|
17
|
+
// V4-Pro 请求按 V4-Pro 自有牌价 9.0/27.0/0.30 计费,不可提前折算。
|
|
13
18
|
// - deepseek-v4-flash-vision-exp 与 deepseek-v4-flash 在 legacy 时代单价一致。
|
|
19
|
+
// - 模型名归一化后 `deepseek-v4.1-flash` 等同为 `deepseek-flash` 的等价写法,命中同一档。
|
|
14
20
|
//
|
|
15
21
|
// 视觉模型 Token 规则:
|
|
16
22
|
// https://api-docs.deepseek.com/zh-cn/guides/vision#token-usage
|
|
@@ -30,14 +36,27 @@
|
|
|
30
36
|
* 均为「高峰时段」价;空闲时段 = 高峰 × 0.5(两版价格的空闲档都恰好是半价)。
|
|
31
37
|
*
|
|
32
38
|
* routes:某个时代内把指定模型名的请求**改按另一档单价计费**,记录也以被路由
|
|
33
|
-
*
|
|
39
|
+
* 到的**计费模型规范名**(当前为官方现役名 `deepseek-flash`)入账,
|
|
40
|
+
* 便于按模型聚合时看到真实计费口径。
|
|
34
41
|
*/
|
|
35
42
|
|
|
36
43
|
/** V4.1 Flash 价格的生效时刻:北京时间 2026-09-10 12:00(UTC+8)= 2026-09-10T04:00:00Z */
|
|
37
44
|
export const V41_EFFECTIVE_AT = Date.UTC(2026, 8, 10, 4, 0, 0)
|
|
38
45
|
|
|
39
|
-
/**
|
|
40
|
-
|
|
46
|
+
/**
|
|
47
|
+
* V4-Pro 请求被路由到 V4.1 Flash 的生效时刻:北京时间 2026-09-14 12:00。
|
|
48
|
+
* 官方表述:「北京时间 2026 年 9 月 14 日 12:00 之后,至未来 V4.1 Pro 上线之前,
|
|
49
|
+
* 用户访问 deepseek-v4-pro 的请求将全部路由到 V4.1 Flash,并按 V4.1 Flash 单价计费」。
|
|
50
|
+
* 注意它与 V41_EFFECTIVE_AT(9-10 12:00)不是同一时刻,提前折算会低估 V4-Pro 花费。
|
|
51
|
+
*/
|
|
52
|
+
export const V41_PRO_ROUTE_AT = Date.UTC(2026, 8, 14, 4, 0, 0)
|
|
53
|
+
|
|
54
|
+
/**
|
|
55
|
+
* V4.1 Flash 档的规范(官方现役)模型名。
|
|
56
|
+
* 官方文档:「模型名请使用 `deepseek-flash`」——被路由的请求一律以此名入账,
|
|
57
|
+
* 使按模型聚合的口径与官方账单一致。`deepseek-v4.1-flash` 归一化后等价命中本档。
|
|
58
|
+
*/
|
|
59
|
+
export const V41_FLASH_MODEL = 'deepseek-flash'
|
|
41
60
|
|
|
42
61
|
export const PRICE_ERAS = [
|
|
43
62
|
{
|
|
@@ -53,14 +72,31 @@ export const PRICE_ERAS = [
|
|
|
53
72
|
},
|
|
54
73
|
{
|
|
55
74
|
id: 'v41',
|
|
56
|
-
label: 'V4.1 Flash 价(
|
|
75
|
+
label: 'V4.1 Flash 价(2026-09-10 12:00 起;V4-Pro 此时仍按自有牌价)',
|
|
57
76
|
since: V41_EFFECTIVE_AT,
|
|
58
77
|
models: {
|
|
59
78
|
// 高峰价:输入(缓存命中)0.04 / 输入(缓存未命中)2 / 输出 8;空闲减半
|
|
60
79
|
[V41_FLASH_MODEL]: { input: 2.0, output: 8.0, cacheRead: 0.04, cacheWrite: 0.04 },
|
|
80
|
+
// V4-Pro 尚在「自有牌价」窗口(9-14 12:00 前不路由),故本时代仍需保留其单价,
|
|
81
|
+
// 否则会落入 provider 兜底而被误标记为「估算」。
|
|
82
|
+
'deepseek-v4-pro': { input: 9.0, output: 27.0, cacheRead: 0.30, cacheWrite: 0.30 },
|
|
61
83
|
},
|
|
62
|
-
//
|
|
63
|
-
//
|
|
84
|
+
// 旧 V4-Flash 系(含视觉版)已下线,请求由 V4.1 Flash 提供并按 Flash 价计费。
|
|
85
|
+
// V4-Pro 此时尚未路由(官方路由时刻为 9-14 12:00,见下一个时代)。
|
|
86
|
+
routes: {
|
|
87
|
+
'deepseek-v4-flash': V41_FLASH_MODEL,
|
|
88
|
+
'deepseek-v4-flash-vision-exp': V41_FLASH_MODEL,
|
|
89
|
+
},
|
|
90
|
+
proRouteSince: V41_PRO_ROUTE_AT,
|
|
91
|
+
},
|
|
92
|
+
{
|
|
93
|
+
id: 'v41pro',
|
|
94
|
+
label: 'V4.1 Flash 价 + V4-Pro 路由(2026-09-14 12:00 起)',
|
|
95
|
+
since: V41_PRO_ROUTE_AT,
|
|
96
|
+
models: {
|
|
97
|
+
[V41_FLASH_MODEL]: { input: 2.0, output: 8.0, cacheRead: 0.04, cacheWrite: 0.04 },
|
|
98
|
+
},
|
|
99
|
+
// V4.1 Pro 上线前,V4-Pro 的请求全部路由到 V4.1 Flash 并按 V4.1 Flash 单价计费。
|
|
64
100
|
routes: {
|
|
65
101
|
'deepseek-v4-pro': V41_FLASH_MODEL,
|
|
66
102
|
'deepseek-v4-flash': V41_FLASH_MODEL,
|
|
@@ -69,7 +105,8 @@ export const PRICE_ERAS = [
|
|
|
69
105
|
},
|
|
70
106
|
]
|
|
71
107
|
|
|
72
|
-
/**
|
|
108
|
+
/** 旧价精确单价表(**仅 legacy 时代**)。保留导出,兼容既有调用与历史口径;
|
|
109
|
+
* 当前时代的精确表请用 `exactModelsAt(ts)`(会随时代切换)。 */
|
|
73
110
|
export const EXACT_MODELS = PRICE_ERAS[0].models
|
|
74
111
|
|
|
75
112
|
/** 归一化模型名:小写并剔除分隔符,使 v4.1 / v4-1 / v41 等写法命中同一档价。 */
|
|
@@ -77,6 +114,18 @@ export function normalizeModelName(m) {
|
|
|
77
114
|
return String(m == null ? '' : m).toLowerCase().replace(/[^a-z0-9]/g, '')
|
|
78
115
|
}
|
|
79
116
|
|
|
117
|
+
/**
|
|
118
|
+
* 规范名别名表(归一化后 → 官方现役规范名)。
|
|
119
|
+
* 官方文档口径:「模型名请使用 `deepseek-flash`」,`deepseek-v4.1-flash` 是同一档价的
|
|
120
|
+
* 等价写法(历史记录与旧文档中出现)。此处**不**把它们加进单价表,以免在单价表展示里
|
|
121
|
+
* 重复列出;只在解析阶段归一到规范名,保证 `estimated=false` 且按模型聚合同一个桶。
|
|
122
|
+
*/
|
|
123
|
+
export const MODEL_ALIASES = {
|
|
124
|
+
deepseekv41flash: V41_FLASH_MODEL, // deepseek-v4.1-flash / deepseek-v41-flash / deepseek_v4.1_flash …
|
|
125
|
+
deepseekv41: V41_FLASH_MODEL, // deepseek-v41
|
|
126
|
+
deepseekflashv41: V41_FLASH_MODEL, // deepseek-flash-v4.1
|
|
127
|
+
}
|
|
128
|
+
|
|
80
129
|
const ERA_INDEX = new Map()
|
|
81
130
|
function eraIndex(era) {
|
|
82
131
|
let idx = ERA_INDEX.get(era)
|
|
@@ -90,13 +139,14 @@ function eraIndex(era) {
|
|
|
90
139
|
}
|
|
91
140
|
|
|
92
141
|
/**
|
|
93
|
-
*
|
|
94
|
-
*
|
|
142
|
+
* 某模型在指定时代下命中的**计费模型规范名**:先按别名表归一,再查本时代单价表,
|
|
143
|
+
* 最后查路由表(路由目标须在本时代单价表内);均未命中返回 null。
|
|
95
144
|
*/
|
|
96
145
|
export function resolveModelInEra(era, model) {
|
|
97
146
|
if (!era) return null
|
|
98
|
-
|
|
147
|
+
let n = normalizeModelName(model)
|
|
99
148
|
if (!n) return null
|
|
149
|
+
if (MODEL_ALIASES[n]) n = normalizeModelName(MODEL_ALIASES[n])
|
|
100
150
|
const idx = eraIndex(era)
|
|
101
151
|
if (idx.models.has(n)) return idx.models.get(n)
|
|
102
152
|
const target = idx.routes.get(n)
|