Skip to content
VECTOR WIREAI INTELLIGENCE
UTC
Home

Qwen4 nears launch with no open weights committed

Fei-Fei Li said on October 29 that Qwen4 could arrive within the week; no weights or license published.

Alibaba Cloud CTO Fei-Fei Li told attendees at the Alibaba Cloud Apsara Conference in Amsterdam on October 29 that Qwen4 could arrive within the week, and a Qwen team researcher reportedly echoed that timeline1. Alibaba Group Holding has separately confirmed that Qwen4 is currently in training2. No release date, model card, license terms, or downloadable weights have been published.

Four variants, no paper trail

Code changes spotted in the Qwen Code repository reference four variant names: qwen4-max, qwen4-plus, qwen4-coder-plus, and hints that the full family will ship a one-million-token context window. A separate analysis identified four named variants in reported materials: Qwen4-Max, Qwen4-Plus, Qwen4-Flash, and Qwen4-27B. Beyond repository artifacts, no technical specifications have been disclosed.

Qwen models have surpassed 700 million downloads on Hugging Face Inc., according to figures reported by China Daily in January. That developer base was built on open weights, and the question of whether Qwen4 preserves that access is now the central concern in community discussion.

A pattern of closing what was once open

Alibaba has said for months that it intends to keep releasing Qwen models under open licenses, and in January a Qwen researcher told Xinhua the team's goal is to keep pushing model performance while staying committed to open-source openness. But recent releases have moved in the opposite direction. Qwen 3.7 never received open weights and remains available only through Alibaba Cloud's API. Qwen3.8 Omni Flash, which launched on September 18, ships API-only with no weights for developers to self-host. Hugging Face's Qwen organization still lists Qwen 3.6 as the newest open release. Alibaba has done closed-then-open before with Qwen2.5 Max, but the recent streak has eroded confidence.

The team itself has also changed. Lin Junyang announced his departure on X on March 4, and Yu Bowen and Hui Binyuan left around the same time. Bloomberg reported that Lin had warned internally about a widening gap with OpenAI before he left. Alibaba never gave a public explanation for the departures.

Separately, The Economic Times carried a report that Alibaba plans to charge large users of its next open-source AI model, though the revenue-sharing rate was described as undecided and the report rested on two anonymous sources.

ANALYSIS The tension is structural: Alibaba's developer ecosystem depends on open weights, but each successive release since Qwen 3.6 has withheld them. Whether Qwen4 ships with downloadable weights or stays API-only will determine whether the 700-million-download community remains on the Qwen stack or begins migrating.

Was this useful?

The day’s vector, in your inbox.

Vector Wire Daily: one dispatch each publishing day.