Skip to content

Alibaba (Qwen)

Dossier checked 28 Sep 2026

What it sells now

Newest frontier release

Qwen3.8-Max-0902

2 Sep 2026 · the lab’s announcement

Upgraded snapshot of Qwen3.8-Max on the international and global API, aimed at long-horizon coding and multi-tool agent work; same price as Qwen3.8-Max

Flagship list price, per million tokens

$2.00 in $6.00 out

qwen3.8-max, international (Singapore) region, released 3 Aug 2026 · pricing page

Qwen3.8-Max-0902, the upgraded snapshot of 2 September, lists the same international price.

Weights

Open flagship

Open, flagship included. Qwen3.8-Max’s text-only weights are public (2.4T parameters, 95B active) under a custom licence that is free to use but needs a separate licence for model-as-a-service or assistant businesses above US$50M a year. Most smaller Qwen models are Apache 2.0. Source

Where you can buy the flagship

Source

11 releases in the year to 30 Oct 2026, 10 the year before · 30 in all since 2023

Our read, not a fact

Split

The question we ask of every lab: is most of its value in the model, which gets cheaper fastest, or in something other work is built on? Our answer: split between the two.

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

01 · What it runs on, and the money

Trains and serves on

Its own cloud: Qwen is built inside Alibaba, so the lab and the datacentre are one company. Its T-Head unit designs accelerators: the Zhenwu chips are used by more than 650 outside customers, and the V900, announced in September 2026, is due in mass production in the first quarter of 2027. What the flagship was trained on has not been disclosed. Source

What it depends on Alibaba designs its own accelerators but has not disclosed what trained its flagship; its capex rose 75% in the June quarter, partly on higher chip-component prices. Source

Biggest risk The segment holding Qwen lost RMB 13.9B at adjusted EBITA in one quarter, partly on the Qwen app’s inference costs, so the open-weight strategy depends on the parent’s cloud and commerce profits. Source

Revenue
In the quarter to 30 Jun 2026, AI Cloud and Compute Services revenue was RMB 48.4B (US$7.1B), up 45%. The new AI Labs and Applications segment, which holds Qwen, had revenue of RMB 3.3B and an adjusted EBITA loss of RMB 13.9B. Source
People
132,165 across Alibaba Group (30 Jun 2026); the Qwen team is not broken out Source
Structure
A programme inside a public company. Alibaba Group is listed in New York and Hong Kong; since the June 2026 quarter the Qwen labs report inside the AI Labs and Applications segment. Source

NotesThe year-ago comparison uses today’s international list price for the Sep 2025 snapshot, because its launch-day international price could not be confirmed; on that basis the flagship’s input price rose and its output price held. Alibaba held about 36% of Moonshot AI per its 2024 filing, likely diluted since.

02 · Releases

data through 30 Oct 2026

  1. Alibaba 27 Qwen3.8-Max-09022 Sep 2026

Each dot is a frontier release at its date. Click one to find it in the full log. Every lab on one chart

The landmarks

  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: the flagship opened to the public on its website and to businesses through the API OpenFirst release
The full release logall 30, newest first
  1. 2026
  2. 2 Sep 2026 Qwen3.8-Max-0902 Frontier Upgraded snapshot of Qwen3.8-Max on the international and global API, aimed at long-horizon coding and multi-tool agent work; same price as Qwen3.8-Max Source
  3. 26 Aug 2026 Qwen3.8-Flash Fast, low-cost Qwen3.8 model on the international and global API; the smallest of Alibaba’s three tiers Source
  4. 3 Aug 2026 Qwen3.8-Max Frontier Max tier, released through the API; open weights of a text-only version followed Source
  5. 19 Jul 2026 Qwen3.8-Max Preview 2.4T-parameter multimodal preview, only inside Alibaba’s coding and agent subscriptions (Token Plan, Qoder, QoderWork), not on the API; Qwen3.8-Max carries the dot Source
  6. Jun 2026 Qwen3.7-Plus Frontier Vision plus deep reasoning with tool invocation
  7. May 2026 Qwen3.7-Max Frontier Proprietary max tier
  8. 14 Apr 2026 Qwen3.6-Max-Preview Frontier Preview of the largest and most capable Qwen3.6 model, text only, on the international API Source
  9. 1 Apr 2026 Qwen3.6-Plus Frontier Plus tier of the Qwen3.6 vision-language series, on the international and global API; Qwen3.6-Flash and two small open models followed Source
  10. Feb 2026 Qwen3.5 Frontier 397B mixture-of-experts flagship
  11. 23 Jan 2026 Qwen3-Max-Thinking Frontier Max-tier reasoning model (qwen3-max-2026-01-23) in Qwen Chat and the paid API Source
  12. 2025
  13. 3 Nov 2025 Qwen3-Max-Thinking Preview Frontier Early preview of Qwen3-Max-Thinking, an intermediate checkpoint, in Qwen Chat and the API Source
  14. Sep 2025 Qwen3-Max Frontier Proprietary max tier of the Qwen3 family
  15. 11 Sep 2025 Qwen3-Next Frontier Ultra-sparse MoE architecture (80B, 3B active), open weights and on the API priced just above Plus Source
  16. 5 Sep 2025 Qwen3-Max-Preview Frontier Preview of the trillion-parameter Max model in Qwen Chat and the paid API; the general release is its own row Source
  17. 25 Jul 2025 Qwen3-235B-A22B-Thinking-2507 Frontier Named update of the open flagship (thinking), open weights and on the API Source
  18. 21 Jul 2025 Qwen3-235B-A22B-Instruct-2507 Frontier Named update of the open flagship (non-thinking), on the API between the Plus and Max prices Source
  19. Apr 2025 Qwen3 Frontier Apache 2.0 family with hybrid thinking
  20. 5 Mar 2025 QwQ-Plus Frontier Reasoning model on the API priced between Plus and Max, with QwQ-32B as open weights Source
  21. 24 Feb 2025 QwQ-Max-Preview Frontier Max-tier reasoning preview, free in Qwen Chat as “Thinking (QwQ)” Source
  22. 28 Jan 2025 Qwen2.5-Max Frontier Large MoE flagship, on the API and Qwen Chat from launch Source
  23. 2024
  24. Nov 2024 QwQ-32B Open reasoning model
  25. Sep 2024 Qwen2.5 Frontier Broad size lineup, widely fine-tuned
  26. Jun 2024 Qwen2 Frontier Second generation: open weights up to 72B, the 72B served on Alibaba’s API at the Plus price
  27. May 2024 Tongyi Qianwen 2.5 (Qwen-Max-0428) Frontier Flagship update behind Tongyi Qianwen 2.5, on the API and the Tongyi apps
  28. 25 Apr 2024 Qwen1.5-110B Frontier Largest Qwen1.5 model, open weights and on the API above the Plus price Source
  29. 4 Feb 2024 Qwen1.5 Frontier Open family up to 72B, on DashScope and together.ai Source
  30. 2023
  31. Dec 2023 Tongyi Qianwen 2.1 Frontier Flagship update (qwen-max-1201) on the API
  32. 30 Nov 2023 Qwen-72B Frontier Largest open Qwen model of its generation, open weights and on Alibaba’s API Source
  33. 31 Oct 2023 Tongyi Qianwen 2.0 Frontier Second-generation flagship, public in China on the website and apps and through the API Source
  34. Sep 2023 Qwen (Tongyi Qianwen) Frontier First public release: the flagship opened to the public on its website and to businesses through the API

03 · When to reach for them

practical read

Open frontier-scale weights
Reach for Qwen3.8-Max’s text-only release to self-host a Max-class model (Aug 2026).
Permissive licence
Pick the Qwen3.6 open models (27B, 35B-A3B) when Apache 2.0 terms matter (Apr 2026).
Default open base
Use Qwen3.5, from 0.8B to 397B parameters under Apache 2.0, as a base to fine-tune (Feb 2026).

04 · Weights, across all ten

the flagship

Open flagship
DeepSeek Mistral AI Alibaba (Qwen) Moonshot AI Zhipu (Z.ai)
Closed flagship, some open
OpenAI Google Meta xAI
Closed
Anthropic

05 · Essays that read Alibaba (Qwen)

from the archive

The Forest Floor Is The Product Most people are optimising for canopy. The work that survives the next five AI releases is built from the forest floor up. The Model Is Not the Moat If frontier capability keeps centralising, the durable edge shifts outward into trust, workflow fit, and the surrounding package. You're Not Comparing Models. You're Comparing Contracts. Agent benchmarks don't measure models. They measure contracts. Two teams running the same model can publish different scores, and both can be honest.

06 · About Alibaba (Qwen)

at a glance

Parent
Alibaba Cloud
HQ
Hangzhou, China
First launch
2023
Also builds
Qwen app, QwenWork, ModelScope, Qwen Cloud
Known for
Most-used open family

Qwen (Tongyi Qianwen) is the model family from Alibaba, headquartered in Hangzhou. First launched in 2023, it is among the most widely used open-weight families, with most releases under Apache 2.0; since August 2026 the flagship’s text-only weights are public under a custom licence.