EN
EnglishEspañol中文PortuguêsFrançaisالعربية日本語한국어

Demand

Monthly PyPI Downloads of Major AI Python SDKs

Monthly PyPI file downloads for the openai, anthropic, google-genai, and legacy google-generativeai projects. The trend signals activity around major AI Python SDKs, but includes automation and dependency traffic and does not measure installations, developers, API use, or market share.

Monthly PyPI Downloads of Major AI Python SDKs

  • openai
  • anthropic
  • google-genai
  • google-generativeai (legacy)
As of 2026-07-01, openai had the most monthly PyPI file downloads among the tracked SDK packages at 401,618,092 downloads.1B100M10MJun '26{"f":[800,420,56,16],"s":[["openai","#10A37F"],["anthropic","#D97757"],["google-genai","#4285F4"],["google-generativeai (legacy)","#AECBFA"]],"p":[["2026-06-01T00:00:00.000Z","Jun '26",56,[[0,"368,967,358 downloads",94.81,null],[1,"158,124,307 downloads",161.78,null],[2,"286,258,727 downloads",114.87,null],[3,"16,737,071 downloads",339.29,null]],null],["2026-07-01T00:00:00.000Z","Jul '26",784,[[0,"401,618,092 downloads",88.11,null],[1,"186,800,706 downloads",148.61,null],[2,"300,124,721 downloads",111.13,null],[3,"16,526,605 downloads",340.29,null]],null]]}openai: 401.6M on Jul '26. anthropic: 186.8M on Jul '26. google-genai: 300.1M on Jul '26. google-generativeai (legacy): 16.5M on Jul '26
SOURCE: PyPI public BigQuery dataset

As of 2026-07-01, openai had the most monthly PyPI file downloads among the tracked SDK packages at 401,618,092 downloads.

Pro API coming soon

Methodology

Query the official `bigquery-public-data.pypi.file_downloads` table, which PyPI documents as a public dataset populated by Linehaul from package download logs. Count rows with `COUNT(*)`, group by the UTC calendar month of `timestamp` and exact `file.project`, and retain only openai, anthropic, google-genai, and google-generativeai.

The query window begins at 2023-01-01 00:00:00 UTC and ends immediately before 2026-08-01 00:00:00 UTC. Because the research date is August 27, 2026, July 2026 is the latest complete month. No partial August activity is included.

All observed download events are counted without filtering by installer, Python version, operating system, country, or detected CI environment. This makes the chart a reproducible total of rows for each exact project name, not an estimate of human-originated downloads.

Download counts are not installations, unique environments, developers, users, or product usage. Automated CI/CD, bots, scanners, mirror refreshes, and repeated downloads can raise counts; pip caches and internal or unofficial mirrors can suppress or otherwise distort PyPI-observed traffic. Package dependencies can also trigger downloads from software whose users did not directly choose the SDK.

Google's packages are not spliced. The legacy google-generativeai series begins when BigQuery first records it in May 2023. The successor google-genai series begins separately in December 2024, consistent with Google's introduction of the new GenAI SDK alongside Gemini 2.0 in late 2024. Google states that the legacy libraries were deprecated as of November 30, 2025.

No estimates, interpolation, rolling-window reconstruction, pypistats short-window backfill, or zero-filling before package availability are used. Every CSV value is an integer returned by the full-history BigQuery aggregation.

Frequently asked questions

What does this chart measure?

It counts package-file download events in PyPI's public logs for the exact projects openai, anthropic, google-genai, and google-generativeai, grouped by UTC calendar month. It does not count installations, environments, users, projects, or model API calls.

Why are Python AI SDK downloads relevant?

Python is widely used to build AI applications, agents, data workflows, and experiments, and official SDKs are a common route to model APIs. Changes in downloads can signal developer-tool activity and migration between SDK generations, but cannot establish provider adoption or usage by unique developers.

How should the download trends be interpreted?

Use each line primarily to follow activity around the same exact package over time. Comparing levels across packages requires caution because package age, dependencies, release cadence, automation, caching, mirrors, and vendor-specific installation patterns can differ substantially.

How are the figures calculated and sourced?

The chart counts rows in the official bigquery-public-data.pypi.file_downloads table, groups them by the UTC month of the timestamp and exact project name, and includes only complete months. No interpolation, zero-filling, smoothing, or conversion to users is applied.

Does one PyPI download equal one SDK user or installation?

No. One person or automated job may generate many package-file requests, while caches or mirrors may support many installations after one PyPI-observed request. CI/CD, bots, scanners, dependencies, repeated downloads, and mirror refreshes can inflate traffic, so the counts are not market share.

Why are google-genai and google-generativeai separate?

They are different PyPI project names and different SDK generations. Google introduced google-genai with Gemini 2.0 in late 2024 and later deprecated the legacy google-generativeai library. Combining them would hide the migration and create an artificial continuous series.

Why does a package line start after January 2023?

Months with no returned rows are not manufactured as zeroes. anthropic first appears in this query in February 2023, google-generativeai in May 2023, and google-genai in December 2024; openai has recorded downloads from the January 2023 start.

Why is August 2026 excluded?

The chart includes only complete UTC calendar months. The data was researched on August 27, 2026, while August was still in progress, so July 2026 is the latest included month.