Alibaba's Qwen team said Sunday that its next flagship model ranks behind only Anthropic's Claude Fable 5 among the frontier systems it compared itself against. The company published no benchmark scores to support that placement. A preview build, Qwen3.8-Max-Preview, went live the same day on Alibaba's Token Plan subscription and on its Qoder and QoderWork coding platforms, announced during the World AI Conference in Shanghai.

Qwen3.8 is "one of the most powerful models available today, comparable to leading frontier AI models, second only to Fable 5," the team wrote in a post on X, according to the South China Morning Post, which Alibaba owns. The post put the model at 2.4 trillion parameters and said open weights were coming "soon."

That is close to everything Alibaba disclosed. The company gave no further technical specifications, which leaves out the activated-parameter count that determines what a model of this size costs to serve, the license the promised weights would carry, and a date for releasing them. No independent evaluations support the ranking, the SCMP reported.

What Changed

What Alibaba published in May

The company handled its previous flagship launch differently. Qwen3.7-Max arrived on May 20 at the Alibaba Cloud Summit with a published benchmark table that included 80.4% on SWE-bench Verified, 69.7 on Terminal-Bench 2.0, 92.4 on GPQA Diamond and 41.4% on Humanity's Last Exam.

Artificial Analysis, an independent benchmarking firm, then scored Qwen3.7-Max at 56.6 on its Intelligence Index, up 4.8 points from Qwen3.6 Max Preview's 51.8. That was the highest placement a Chinese model had reached on the index at the time. Buyers could hold the vendor's table against a third party's measurement and see where the two agreed.

Sunday's announcement gives them a ranking and a parameter count. Alibaba has not said when a full launch post for Qwen3.8 will follow.

Moonshot published scores three days earlier

The claim landed in a week when a direct competitor did show its work. Moonshot AI unveiled Kimi K3 on Thursday at 2.8 trillion parameters, which the SCMP described as currently the largest open-source model, and released benchmark results with it.

Those results put K3 ahead of Fable 5 and OpenAI's GPT-5.6 Sol on some tests, including Program Bench and SWE Marathon. Arena.AI, formerly LMArena, ranked K3 top in its Frontend Code Arena across five domains, with Fable 5 leading only in Gaming. Moonshot's figures are self-reported, and the company released them with the model.

The open-weight promise runs against Alibaba's recent record

Releasing Qwen3.8's weights would reverse how Alibaba has shipped its two most recent flagships. Qwen3.6-Max-Preview in April and Qwen3.7-Max in May both went out proprietary and API-only, hosted through Alibaba Cloud rather than distributed as downloadable files. The company has not published license terms for Qwen3.8 or said which of the two paths it will take.

Shuai Bai, a developer on the Qwen team, said on X that Qwen3.8 is the group's first multimodal model at this scale, able to process images, video and documents. Access during the preview runs at 10% of the standard price, a discount on a model with no published evaluation results.

What is still missing

Alibaba has published no model card for Qwen3.8 and no license for the weights it promised. The model has not appeared on an independent leaderboard, which is where the comparison against Fable 5 gets tested rather than asserted.

Qwen3.7-Max reached the Artificial Analysis index within days of its May launch. Alibaba has not said when Qwen3.8 will follow it there, or when the weights ship.

Know someone who'd find this useful? ✉️ Email it to a friend in one click, or they can subscribe free here.

Frequently Asked Questions

What did Alibaba actually publish about Qwen3.8?

A parameter count of 2.4 trillion, a ranking claim placing it second only to Claude Fable 5, and a preview build on three of its own platforms. The company gave no benchmark scores, no activated-parameter count, no license for the promised weights and no release date.

Can the "second only to Fable 5" claim be verified?

Not yet. No independent evaluations support the ranking, according to the South China Morning Post. Verification requires the model to appear on an independent leaderboard such as the Artificial Analysis Intelligence Index, where Qwen3.7-Max was scored after its May launch.

How is this different from how Alibaba launched Qwen3.7-Max?

Qwen3.7-Max arrived on May 20 with a published benchmark table covering SWE-bench Verified, Terminal-Bench 2.0, GPQA Diamond and Humanity's Last Exam. Artificial Analysis then scored it at 56.6 on its Intelligence Index, the highest placement a Chinese model had reached at that point.

Will Qwen3.8 really be released as open weights?

Alibaba says so but has set no date and published no license. That commitment would reverse its recent practice, since Qwen3.6-Max-Preview in April and Qwen3.7-Max in May both shipped proprietary and API-only through Alibaba Cloud.

Moonshot Launches Kimi K3 With 2.8 Trillion Parameters and 1M Context
Moonshot AI said Thursday that it had launched Kimi K3, a model built with 2.8 trillion total parameters. K3 is available now through Moonshot's products and API. Full weights remain unavailable; Moon
The Implicator
GLM-5.2 Becomes the Top Open-Weight Model on Artificial Analysis
Artificial Analysis this week named Z.ai's GLM-5.2 the leading open-weight model on its Intelligence Index, the independent benchmarking firm said in a post on June 16. The model scored 51 on the v4.1
The Implicator
AI News

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai