DeepSeek Harness Metrics

The official site publishes a few points it chooses to publish. We list three sets of figures under three definitions — never added together, never smoothed over, every number carrying its source.

Why three groups. On 5 November the official team disclosed 1.54M DAU, 92 trillion tokens/day and ~35k plugins. But only three time points exist; there is no continuous series.

So we add two more:
· What we collect: GitHub stars, weekly npm downloads, OpenRouter tokens — daily
· What research firms estimate: Counterpoint, Presenc AI and others — with source and confidence

The three definitions are not additive and not directly comparable. Anyone who merges them produces a false number.
DSH GitHub Stars
274,494
DSH Forks
33,536
npm 週下載
826,814
官方披露 DAU
154 萬

1. Official disclosure (verified)

Source: official release of 5 November, via IT之家 and 21st Century Business Herald / The Paper. DAU is a 7-day rolling average over users of the official API.

DSH daily active users (official)

Three official points only. DAU is a 7-day rolling average over official-API users.

182 万136 万90.9 万45.4 万0.0 万27.1 万09-3042.9 万10-07154 万11-04DAU(万)

來源:官方披露(11/5 晚),經 21 世紀經濟報道/澎湃新聞轉述

DateValue
2026-09-3027.1 万
2026-10-0742.9 万
2026-11-04154 万

DSH daily token consumption (official)

One official point: ~92 trillion on 2026-11-04.

Only one data point so far; the chart appears once there are two.

DateValue
2026-11-0492 万亿

Third-party plugin count (official)

Officially ~35,000 as of 11/5. Our directory lists 8,560.

Only one data point so far; the chart appears once there are two.

DateValue
2026-11-053.5 万

2. Collected by us (continuous, verifiable)

Not published by the official site, but all publicly queryable. Collected daily.

DSH GitHub stars (collected daily by us)

The official site does not publish this, but it is a public, verifiable attention signal. Collected daily.

Only one data point so far; the chart appears once there are two.

DateValue
2026-11-06274,494

DSH weekly npm downloads (collected daily by us)

Weekly downloads of @deepseek-ai/dsh — a direct signal of how many people install it.

Only one data point so far; the chart appears once there are two.

DateValue
2026-11-06826,814

Weekly token volume, whole market (OpenRouter / Sacra)

All of OpenRouter's weekly volume, not all of DeepSeek.

41.8 T31.3 T20.9 T10.4 T0.0 T0.3 T01-0510.5 T07-2717.3 T08-0333.6 T10-1735.4 T10-26萬億 token/週

來源:OpenRouter / Sacra

DateValue
2026-01-050.3 T
2026-07-2710.5 T
2026-08-0317.3 T
2026-10-1733.6 T
2026-10-2635.4 T

OpenRouter daily total tokens (collected by us)

Collected daily at 07:20. OpenRouter channel only; not comparable with the official figure.

10.3 T7.8 T5.2 T2.6 T0.0 T8.8 T11-058.0 T11-06萬億 token/日

來源:openrouter.ai/rankings(每日採集)

DateValue
2026-11-058.8 T
2026-11-068.0 T

DeepSeek Harness weekly tokens on OpenRouter

The DSH app alone, on a third-party channel.

2.6 T2.0 T1.3 T0.7 T0.0 T2.2 T11-052.2 T11-06萬億 token/週

來源:openrouter.ai/rankings(每日採集)

DateValue
2026-11-052.2 T
2026-11-062.2 T

3. Third-party estimates (not independently verified)

How to read this table. These are estimates from research firms using different definitions (app, web, all-platform). They cannot be summed, nor compared directly with the official figure.
We list them because how each party counts is itself informative — not because they are right.
SourceMetricValueAs ofScopeConfidence
Counterpoint Research全球月活(MAU)約 1.45 億2026-04DeepSeek App 全球中
Presenc AI週活(WAU)+ 地域分佈1.88 億2026-Q1中國 47% / 美國 11% / 東南亞 14% / 印度 9% / 歐洲 8%低
Search Engine Land全球月活(MAU)9,688 萬2025-04DeepSeek 網頁端中
搜狐 / 多篇報導網頁端用戶來源中國 33.5% / 俄羅斯 7.1% / 美國 6.6%2026-03DeepSeek 網頁端低

4. Why bother with this "boring" work

The official site will not publish GitHub stars, npm downloads or OpenRouter app-level tokens — yet those are the most direct signals of real ecosystem heat.
Research firms publish reports with inconsistent definitions and mostly undisclosed methods.
Recording each one, citing it, and keeping the definitions apart — nobody does that. Do it long enough and it becomes a history that exists nowhere else.

Related

User scale and definitions · Token trend · Ecosystem events · This week · Benchmarks