Google ADK leads recent Python downloads with 28.6M in 30 days — but none of the four frameworks publishes a true active-user count
Asked:
“Active users report for Frameworks Langchain vs Pydantic AI vs Microsoft Agentic Framework vs Google ADK”
Scanned 2026-09-01 across official GitHub repositories, PyPI Stats, Pepy and independent web sources. The fairest cross-framework activity proxy available is downloads of each framework’s primary Python package over the last 30 days; three of the four packages exposed a recent count, LangChain’s did not.
Last-30-day PyPI downloads of the primary package
Bars: downloads in the 30 days to 2026-09-01. Downloads are an activity proxy — not monthly active users. LangChain is shown as unresolved, not zero.
What the chart cannot say
Google ADK holds 60.9% of the measured three-way recent-download total; Pydantic AI 34.3%, Microsoft Agent Framework 4.8% —
pepy.techPydantic AI has the largest all-time base of the three measured packages at 187.9M cumulative downloads, versus 62.4M for google-adk and 11.1M for agent-framework-core —
pepy.techThe exact langchain package shows 2.13B cumulative downloads, but the fetched page exposed no last-30-day figure, so LangChain is left unresolved rather than ranked —
pepy.tech“Microsoft Agentic Framework” is read as Microsoft Agent Framework, whose core Python package is agent-framework-core —
github.comMethod and limitations
No apples-to-apples active-user data is public for any of these frameworks. Package downloads are the fallback proxy, and they distort:
- CI pipelines, caches, mirrors, bots and reinstalls inflate counts without adding users.
- Dependency resolution counts a download every time another package pulls the framework in.
- Package age and modular packaging differ: LangChain splits across many packages (langchain-core alone shows 1.9B cumulative), while newer frameworks concentrate in one.
- Cumulative totals reward age, so they are never mixed into the recent-activity ranking.
- Fetched official GitHub pages exposed no quantitative counters in parsable text; PyPI Stats pages likewise exposed no numbers.
Do not read any figure on this page as monthly active users.
The rows
| Rank | Framework | Package | 30-day downloads | Avg / day | All-time | Share of measured | Source |
|---|
One comparison row per framework, 4 rows; supporting metrics from 37 discovered public package and repository signals plus 12 direct page fetches. Checked 2026-09-01. A downloads figure counts PyPI package pulls of the named package in the stated window, not unique users. Adjacent ecosystem packages (extensions, broad Pydantic packages) were excluded from the comparison for fairness.