Preface¶

Starting from 00:00 (Beijing Time) on August 17, 2026, DeepSeek API will adopt tiered pricing based on peak/off-peak periods. Peak hours are 9:00–12:00 and 14:00–18:00 Beijing Time, and off-peak prices are half of peak prices. For users running agents in DeepSeek Harness (DSH), this means the unit price for the same conversation and same model can double between 10 AM and 10 PM.

The core philosophy of DSH is “everything is a plugin”: the web interface, tools, and themes can all be added via community plugins. The official repository is deepseek-ai/deepseek-harness. There is an independent community plugin directory site for searching and installing these plugins, which has no official affiliation with DeepSeek/HyperGAI and should not be treated as an official app store.

The default web interface does not display the current API unit price in the session top bar. Without reminders during peak hours, it is easy to continue running long tasks based on the impression of a fixed one-time price. dsh-client-pricing does a very specific thing: pins the currently active input, output, and cache hit prices to the session top bar, and switches dynamically with flash/pro tiers and peak/off-peak periods.

This article is collated after cross-checking the community directory details page, the GitHub repository README, and DeepSeek’s official “Models & Pricing” documentation.

What is this¶

dsh-client-pricing is a web client plugin for DeepSeek Harness, maintained by Miyazawai, licensed under MIT, and primarily written in TypeScript. The community directory categorizes it under “Tools & Capabilities”. The GitHub repository currently has 13 stars (the directory page shows 8 stars at the time of inclusion; refer to real-time repository data for the latest count).

It does not solve session billing statistics or register pricing query tools for agents, but a narrower task: real-time display of the currently active DeepSeek API unit price (unit: yuan / 1M tokens) in the session top bar, to avoid you estimating costs based on the old fixed price during peak hours.

The price numbers come from a hard-coded price list in the repository, cross-referenced with the official Models & Pricing documentation. The official page is a static document with no pricing API, so the plugin does not scrape prices from the web in real time.

Core Features¶

Top Bar Tri-State Badge and Three Prices¶

After installing and restarting the web client, a new horizontally centered, single-line, non-wrapping price bar will appear in the session top bar:
- Before tiered pricing takes effect: The badge shows “Standard Fixed Price” in neutral gray.
- After activation · Peak: The badge shows “Peak” in orange.
- After activation · Off-peak: The badge shows “Off-Peak” in green.

Next to the badge are the three currently active numbers: input (cache miss), output, and cache hit. Starting on the day this article was written (2026-08-17), the official tiered pricing has taken effect, so you should normally see either “Peak” or “Off-Peak” instead of the fixed price. The repository still retains the pre-activation fixed price list to make this transition logic testable and traceable.

Hover Overlay¶

When you hover your mouse over the bar, a full price list for the current model will pop up, showing peak/off-peak tiers multiplied by three columns: cache hit, cache miss, and output, along with a description of the period rules. Before tiered pricing takes effect, the overlay will also preview the two new price tiers and their activation time.

Auto-Switch Based on Model and Period¶

• The plugin subscribes to modelDirectories. When the current session switches between flash and pro, the top bar numbers will immediately update to the corresponding tiers.
• The model name mapping is hardcoded as: deepseek-v4-flash → flash, deepseek-v4-pro → pro. Other models will be automatically hidden, leaving no blank price tag in the top bar.
• The tier is recalculated every 30 seconds. When crossing boundaries such as 9:00, 12:00, 14:00, or 18:00, the badge and numbers will update automatically without refreshing the page.
• The time period is calculated according to Asia/Shanghai (UTC+8, no daylight saving time), and does not follow the browser’s local time zone. The window is left-closed and right-open: peak hours start at 9:00 and end at 12:00, then resume from 14:00 to 18:00.

Built-in Price List (Cross-Checked Against Official Documentation)¶

The numbers given in the repository README match the official documentation as of 2026-08-17, with units all in yuan / 1M tokens:

There are two official pricing rules to remember: off-peak price = peak price × 0.5; peak hours are 9:00–12:00 and 14:00–18:00 Beijing Time, and all other times are off-peak.

Installation and Activation¶

The installation command given on the community directory page is as follows, run it in the DeepSeek Harness terminal:

dsh plugin add github:Miyazawai/dsh-client-pricing

This is the original text from the directory page; do not modify the owner/repo combination yourself. For reproducible installations, fix the commit hash as specified on the directory page:

dsh plugin add github:Miyazawai/dsh-client-pricing#commit

Replace #commit with the actual commit SHA from the repository.

The repository README adds details about its use as a web profile client plugin. The equivalent profile command is:

dsh plugin --profile web add dsh-client-pricing

The README also provides an install.ps1 one-click script (which automatically appends the loader line), as well as manual installation steps:
1. Navigate to the web profile directory to install the package: cd ~/.dsh/profiles/web, then run corepack pnpm add dsh-client-pricing.
2. Append the loader line to cordis.patch.yml (skip if it already exists):

- insert:
    - id: dsh-client-pricing
      name: 'dsh-client-pricing'

You must restart dsh web after adding the new loader line. The reason is that ClientModuleRegistry permanently caches package metadata: stop the current process (Ctrl+C, or end the process occupying port 3080), then run dsh web, and refresh http://127.0.0.1:3080. Afterwards, if you only modify the plugin source code, HMR will hot-update after about 500ms after running npm run build, and no restart is needed.

To uninstall, run corepack pnpm remove dsh-client-pricing in the same profile directory, delete the corresponding line from cordis.patch.yml, and restart dsh web again.

Typical Usage¶

After installing and restarting, open any session in the web interface that uses deepseek-v4-flash or deepseek-v4-pro, and check the top bar.
1. Confirm that the current model is flash or pro. The price bar will disappear when you switch to other models, which is expected behavior.
2. Check the badge color: orange means peak hours, green means off-peak hours. Compare with whether the current Beijing Time falls within 9:00–12:00 or 14:00–18:00.
3. Read the three numbers. For example, flash during peak hours: cache miss 3.0, output 9.0, cache hit 0.10.
4. Hover your mouse to check the complete peak/off-peak price list for the model and the period description.
5. Switch the model in the session from flash to pro, and the top bar should immediately update to the three prices for pro (cache miss 9.0, output 27.0, cache hit 0.30 during peak hours).

It will not estimate the current round of costs below the input box, nor will it modify agent behavior. The top bar is only a unit price dashboard.

If the official adjusts prices in the future, you will need to modify src/pricing.ts in the repository: the legacy fixed price is in LEGACY_PRICES, peak prices are in PEAK_PRICES (off-peak prices are automatically derived by multiplying peak prices by 0.5), the activation time is in NEW_PRICING_UTC_MS, and peak windows are in PEAK_WINDOWS. After making changes, run npm test and npm run build. The README includes these four constants and testing conventions in a maintenance prompt that can be directly handed to coding assistants to update prices based on the official pricing page.

Applicable Scenarios and Notes¶

Who this is for: Users who already use the DSH web interface, mainly run official DeepSeek flash/pro models, and want to quickly see “whether it is currently peak or off-peak”. It is more convenient than checking the official documentation for long sessions, rushing work during peak hours, or needing to explain to colleagues “why the same task is more expensive this morning”.

Who this is not for:
- Users who need to calculate the cost of the current session based on token usage, daily totals, or register price_estimate-style tools for agents. This plugin does not handle cost statistics.
- Users using models other than deepseek-v4-flash / deepseek-v4-pro. The price bar will be hidden.
- Users who want prices to automatically sync with the official website in real time. The price list is hard-coded, so you will need to wait for a repository update or modify src/pricing.ts yourself after the official website changes prices.

Important notes before use:
- The plugin runs with the permissions of the current dsh process, and may execute code during installation. You should check the source code repository and license before installing.
- The overlay uses position:absolute centered relative to the header. It may overlap with elements on both sides in extremely narrow windows; the README notes that this is purely for display and does not block interactions.
- The community directory is an independent site. Follow the original text on the directory page for installation commands, and do not mix up this plugin with third-party “pricing” plugins with the same name.

Summary¶

dsh-client-pricing places the current unit price of DeepSeek API on the DSH session top bar: two tiers for peak/off-peak, automatic price switching for flash/pro, and recalculates tiers every 30 seconds. As of 2026-08-17, the official tiered pricing has taken effect, with peak prices double off-peak prices; having a top bar reminder is more reliable than judging the period based on memory.

Community Directory: https://deepseek-harness-plugin.com/en-US/plugins/dsh-client-pricing/

GitHub: https://github.com/Miyazawai/dsh-client-pricing