Alibaba's Qwen reaches 2.4T parameters, Qwen 4 in training
Alibaba’s Qwen model family has evolved from a 7-billion-parameter open-source release in 2023 to a 2.4-trillion-parameter flagship by August 2026. The latest open-weight model, Qwen3.8-2.4T-A95B, features 95 billion active parameters and introduces a commercial license requirement for providers earning over US$50 million annually. This release follows a rapid cadence in 2026, where Alibaba…
Key points
- Alibaba open-sourced Qwen3.8-2.4T-A95B with 2.4T parameters and 95B active parameters in August 2026.
- Qwen 4 is currently in training, with Qwen 4.5 and 5 projected at 5 to 10 trillion parameters.
- Alibaba released four Qwen generations in six months in 2026, splitting open and proprietary strategies.
The ecosystem has expanded significantly, with over 400 open models downloaded more than 1 billion times and the Qwen app reaching 203 million monthly active users. However, the team faced internal turbulence in March 2026, with three senior figures leaving, including tech lead Junyang Lin. In response, Alibaba reorganized its AI units under CEO Eddie Wu. Looking ahead, Alibaba announced at Apsara 2026 that Qwen 4 is currently in training, with a roadmap projecting Qwen 4.5 and Qwen 5 to reach 5 to 10 trillion parameters, though no specific release dates or licenses have been confirmed for these future iterations.
Model page: Qwen3.8-2.4T-A95B →
The story so far
8 episodes →- Alibaba's Qwen reaches 2.4T parameters, Qwen 4 in trainingthis story
The Story of Qwen: Alibaba’s AI Models From 7B to 2.4T
MarkTechPost · 5 October 2026
Loading the full article…
This text was published by MarkTechPost and written by Asif Razzaq. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- Reflection AI prepares to release its first open-weight model to rival DeepSeek and Qwen · 3 src
- Claude Opus 5.5 tested for modeling a castle town · 1 src
- OpenAI, Google, and Anthropic launch new AI models in late September · 3 src
- Anthropic releases Claude Sonnet 5.5 with 30% speed and cost gains · 5 src
- Anthropic releases Opus 5.5 and Sonnet 5.5; OpenAI launches GPT-6 Sol and Luna tiers · 5 src
Comments
via GitHub Discussions