Market

Alibaba’s Qwen 4 reaches the API with closed weights for now, but the reported size and Artificial Analysis score differ from the figures circulating for “Qwen4-Max”

Alibaba’s Qwen 4 is API-only for now, with open weights promised. One source gives 1.2T parameters and an index score of 58, not the 2.4T and 43 in circulation.

By ·

Alibaba’s Qwen 4 launched through its API on November 11, 2026, with weights described as closed for now. The figures in circulation for “Qwen4-Max” (2.4 trillion parameters, a score of 43 on the Artificial Analysis Index, behind Kimi K3.1 Pro) do not match what we found: one technical write-up gives about 1.2 trillion total parameters and a score of 58, three points above Kimi K3.1 Pro at 55. The 2.4 trillion figure matches the earlier Qwen3.8-Max. We could not reconcile the two sets of numbers.

The essentials

Two sets of numbers

The 2.4 trillion total and 95 billion active parameters match Qwen3.8-Max, released in August, which Artificial Analysis scored 58 on an earlier index version. A different index version can give a very different number, which may explain a score of 43. We cannot confirm that, and the “Max” label for Qwen 4 appears in an Apsara conference report that says the models were shown before release.

Caveats

Most of the benchmark figures come from Alibaba or from one secondary write-up that says it has not reproduced them. Open-weights promises have a date but no delivery yet. We read one write-up directly and saw the others only as search excerpts.

What is not established

Summary table

Summary table
PointWhat is stated as of November 7, 2026Source
LaunchAPI live November 11, 2026, weights closed for nowlocal-ai-zone
Reported sizeAbout 1.2T total, 56B active, 1M contextlocal-ai-zone
Index score58 on Artificial Analysis Index v5, Kimi K3.1 Pro 55local-ai-zone
Apsara previewLineup shown before release, per the reportOrcaRouter
Earlier flagshipQwen3.8-Max, 58 on an earlier index versionDeepLearning.AI

FAQ

Is Qwen4-Max open source?

Not yet: the API is live, and Apache 2.0 weights are promised by December 20, according to the write-up we read.

Does it have 2.4 trillion parameters?

The 2.4 trillion figure matches Qwen3.8-Max; the Qwen 4 write-up gives about 1.2 trillion.

Did it score 43?

We found 58 on Index v5 in one source. We could not find a 43 or tell which index version it would refer to.

Method and limits

Checked on November 16, 2026. We read the local-ai-zone write-up through a page summary; OrcaRouter and DeepLearning.AI were seen as search excerpts. We found no Alibaba primary page.

Sources

Spotted an error in this article? Report it · Our corrections policy

OrigoTech logo

New: the app iPhone & iPad

OrigoTech is now an iPhone and iPad app

Read the articles natively, with “New” markers on unread ones, local search and offline reading of what you have already loaded. Free, no account, iOS 17 or later; ads appear between articles.

Rekha icon: a house drawn in a single stroke

From the same author Self-promotion

An Eulerian path, drawn with your finger

Rekha, the iPhone game by OrigoTech's author: connect every dot in a single stroke without tracing the same link twice, across 500 drawings that are all solvable.

The OrigoTech newsletter

A sourced weekly digest of AI tool news.

Read next