12 models · on the curve since 2023

Alibaba (Qwen)

Value position · the site’s read Mixed

Qwen’s durable position is ubiquity: the most downloaded open-weight family, the default base thousands of teams fine-tune, wired into Alibaba Cloud’s distribution across Asia. The models commoditise; the ecosystem around them is the moat.

What would change this readAnother open family displaces Qwen as the default base to fine-tune, or Alibaba Cloud loses its regional distribution edge.

Recent cadence
6 in last 12 mo
4 the year before
Landmark density
4 / 12
defining releases
Weights
Open + closed
Qwen3.7-Max flagship

Founded Alibaba CloudHangzhou, China

The substrate underneath figures as of Aug 2026

Trains and serves on

Its own cloud — Qwen is built inside Alibaba Cloud Intelligence, so the lab and the datacentre are the same company. In May 2026 its T-Head unit unveiled the Zhenwu M890, an in-house accelerator with 144GB of HBM3 on a domestic 7nm process, and demonstrated Qwen3.7-Max running on it. What the flagship was TRAINED on has never been disclosed.

ConcentrationAlone among these labs it owns the whole stack — silicon design, datacentres, cloud and model — yet its frontier training capacity still rests on NVIDIA parts gated by both US export policy and Chinese import approval.

Flagship, per million tokens

$2.50 in $7.50 out

qwen3.7-max, international region

Was

$0.86 / $3.44

Qwen3-Max at launch · Sep 2025

Weights
Split, and the split is the story. The open tier is genuinely Apache-2.0 down to 35B mixture-of-experts models with very long context. The flagship is not: qwen3.7-max has no published weights and is sold only through Alibaba Cloud.
Where you buy it
Model Studio and Bailian · HuggingFace · ModelScope · AWS Bedrock · Microsoft Foundry · OpenRouter
Structure
A programme inside a public company, not a startup. Alibaba Group is dual-primary listed on the New York and Hong Kong exchanges, so Qwen carries no separate funding or valuation of its own.
Revenue
Cloud Intelligence revenue was RMB 41.6B (US$6.0B) for the quarter to 31 Mar 2026, up 38%. AI-related product revenue reached RMB 9.0B in that quarter, about 30% of external cloud revenue, annualising past RMB 35.8B (~US$5.2B). Qwen itself is not broken out.
People
126,661 across Alibaba Group (30 Sep 2025); the Qwen team is not disclosed separately

Biggest single dependencyIts flagship is closed-weight and monetised solely through its own cloud, which puts revenue, compute supply and chip roadmap on one balance sheet exposed to a single import-approval regime.

No 2024 flagship price could be verified from a primary source, so the comparison above starts at Sep 2025 — on that basis the flagship line has moved UP while the previous generation was cut by about half. Alibaba holds roughly 36% of Moonshot AI per its own 2024 filing, near-certainly diluted since. The reported Qwen3.8-Max preview appears only in aggregator sources and is not confirmed by Alibaba.

On the durability curve frontier era
2023202420252026 Qwen (Tongyi Qianwen)Qwen3.8-Max Preview ARCHITECTURE the substrate · compounds  → OUTPUT per model · commoditises  → CROSSOVER the durability thesis  → DURABLE VALUE-CAPTURE ↑ the margin a release keeps · not raw capability TIME · FRONTIER MODEL RELEASES →
Output · value a model keeps Architecture · the compounding substrate Alibaba (Qwen) release

Hover any point to read the release. The y-axis is qualitative; only the x-axis carries dates.

When to reach for them practical read
Default open base
Reach for Qwen2.5 as a widely-supported base to fine-tune (Sep 2024).
Permissive license
Pick Qwen3 when Apache 2.0 terms matter for shipping (Apr 2025).
Open reasoning
Use QwQ for open reasoning without a closed API (Nov 2024).
The landmark releases 4 of 12 models
  1. Feb 2026 Qwen3.5 397B mixture-of-experts flagship MoEFlagship
  2. Apr 2025 Qwen3 Apache 2.0 family with hybrid thinking OpenApache 2.0
  3. Sep 2024 Qwen2.5 Broad size lineup, widely fine-tuned OpenMany sizes
  4. Sep 2023 Qwen (Tongyi Qianwen) First public release OpenFirst release
Full release logall 12, newest first
  1. 2026
  2. Jul Qwen3.8-Max Preview 2.4T-parameter multimodal, previewed at WAIC — a preview, not general availability Preview
  3. Jun Qwen3.7-Plus Vision plus deep reasoning with tool invocation VisionReasoning
  4. May Qwen3.7-Max Proprietary max tier Max tier
  5. Apr Qwen3.6 / Omni Mixed open and proprietary, any-modality OmniMixed weights
  6. Feb Qwen3.5 397B mixture-of-experts flagship MoEFlagship
  7. 2025
  8. Sep Qwen3-Next Ultra-sparse MoE architecture OpenSparse MoE
  9. Apr Qwen3 Apache 2.0 family with hybrid thinking OpenApache 2.0
  10. Jan Qwen2.5-Max Large MoE flagship MoEFlagship
  11. 2024
  12. Nov QwQ-32B Open reasoning model OpenReasoning
  13. Sep Qwen2.5 Broad size lineup, widely fine-tuned OpenMany sizes
  14. Jun Qwen2 Second generation Open
  15. 2023
  16. Sep Qwen (Tongyi Qianwen) First public release OpenFirst release
Where they sit on weights versus the field
The durability take on Alibaba (Qwen) from the archive
About Alibaba (Qwen) at a glance
Parent
Alibaba Cloud
HQ
Hangzhou, China
First launch
2023
Flagship
Qwen3.7-Max
Weights
Mixed: open tier, closed flagship
Also builds
Tongyi, ModelScope
Known for
Most-used open family

Qwen (Tongyi Qianwen) is the model family from Alibaba Cloud, headquartered in Hangzhou. First launched in 2023, it is among the most widely used open-weight families globally, with most releases under the permissive Apache 2.0 license alongside proprietary flagships.

The frontier chart → Read the archive →