Two numbers, two rulers
Set the claims side by side and the disagreement mostly dissolves. They are not measuring the same thing.
| Alibaba's statement | Hugging Face's report | |
|---|---|---|
| Downloads | More than 3 billion | 2,045M (2,061M counting all repositories) |
| Window | The past six months | First seven months of 2026 |
| Scope | Global, no breakdown given | Activity on the Hugging Face Hub |
| Derivatives | More than 300,000 | 151,448 on the Hub |
| Models published | More than 460 | Not stated |
A global count sitting above a Hub-only count is what you would expect, and the two derivative counts split along the same line. Which other distribution channels Alibaba folded into its total is not stated in either source, so the size of the gap cannot be attributed. Neither figure is wrong; they answer different questions, and only one of them was published with its method attached.
"Alibaba Group Holding’s open-weight models have accumulated more than 3 billion global downloads in the past six months"／"Google, part of Alphabet, had 418 million downloads while Meta stood at 227 million in 2026, according to popular open-source AI hub Hugging Face Inc."／"Qwen, Alibaba’s family of AI models, has open-sourced more than 460 models"／"its ecosystem has spawned 300,000-plus derivatives"— from Fortune's report. The first, third and fourth fragments report Alibaba's own statement; the second is Fortune's attribution of the Google and Meta figures to Hugging Face
What the report says about Qwen downloads
Checked against the live page on August 17, 2026: the strings 3 billion, 418, 227, 300,000 and 460 do not appear anywhere in Hugging Face's State of Open Models: Summer 2026. The comparison being passed around — Qwen 3B against Google 418M against Meta 227M — is therefore not a table lifted from that report, whatever the attribution in a given writeup says. Fortune's own wording credits the Google and Meta figures to Hugging Face without pointing at the post, which is likely a data pull rather than a citation.
The report states its own scope in one sentence, in a Notes section: the analysis covers activity observed on the Hugging Face Hub during the first seven months of 2026. That boundary is what makes 3 billion incomparable rather than wrong — a different platform and a different clock. Inside it, Qwen's total is 2,045M across repositories with declared parameter counts, or 2,061M counting every repository.
Long data posts like this one put that caveat far below the numbers everyone screenshots. Pulling the page into markdown keeps the caveat next to the figure it governs.
"This analysis is based on activity observed on the Hugging Face Hub during the first seven months of 2026." (Notes on method)／"Moonshot’s frontier-only portfolio recorded 37M downloads over the year, while Qwen’s broader release strategy across model sizes reached 2,045M (across repositories with declared parameter counts, 2,061M including all repositories) , about 55 times more." (2. Attention ≠ Adoption)— from Hugging Face's published report
The lead survives the correction
None of this dents the underlying claim. On Hugging Face's own published data, Qwen is the base model the community builds on: 151,448 derivatives, which the report puts at 2.6 times Meta's total footprint and 4.7 times the Llama repositories specifically, with Google second at 82,506. The growth is steady rather than launch-driven, running at roughly 180 to 210 new repositories a day through the first seven months of 2026 — which the report reads as adoption that is not driven only by individual launches, and states plainly: Qwen has become part of the default workflow for developers deciding what to fine-tune and deploy. That is a claim about position rather than about volume, and it does not depend on which download total you accept.
Where the volume sits is the more useful correction to the headline. Among models declaring a parameter count, everything under 1B takes 83% of all-time downloads while everything above 100B takes 1%. A download lead is therefore largely a lead in small models that run on ordinary hardware. For the full landscape read the state of open models for summer 2026, and for the model line itself, Qwen3.8.
"Qwen-based models now account for 151,448 derivatives on the Hub, 2.6× Meta’s total footprint and 4.7× the Llama repositories specifically. Google follows with 82,506 derivatives."／"Qwen derivatives have increased at roughly 180–210 new repositories per day throughout the first seven months of 2026, showing that adoption is not driven only by individual launches. Qwen has become part of the default workflow for developers deciding what models to fine-tune and deploy."／"Among models that declare a parameter count, those under 1B take 83% of all-time downloads and everything above 100B takes 1%."— from Hugging Face's published report
How to quote either number
If you are citing 3 billion, say what it is: Alibaba's own six-month, all-platform count, given to reporters. If you are citing 2,045 million, say that too: Hugging Face Hub activity over the first seven months of 2026. Neither belongs in the same row of a table as the other, and adding them together produces nothing. The distinction matters most for the derivative counts, where 300,000 and 151,448 are the same measurement taken with two different definitions of where Qwen lives.
