
Qwen3.8 Max preview lands with 2.4T parameters, but proof is still to come
Alibaba's Qwen3.8-Max-Preview is a 2.4T-parameter model available through Qoder and Token Plan, but its benchmark evidence and open-weight release are still pending.
What launched
Alibaba's Qwen team announced Qwen3.8-Max-Preview on July 19 as a 2.4-trillion-parameter flagship foundation model. Qwen said the full Qwen3.8 is going open-weight soon and described the model as comparable to leading frontier systems, second only to Anthropic's Fable 5. Those are the lab's claims, not an independently verified ranking. 1 2
The preview is already available through Alibaba's Token Plan, Qoder, and QoderWork. Alibaba Cloud's model directory currently labels
qwen3.8-max-preview as available only to Token Plan subscribers, while Qoder's launch page lists Qoder Desktop, JetBrains, CLI, cloud agents, and mobile/web products as supported entry points. 1 3 4What is new
Qoder says the new model improves on Qwen3.7-Max in coding and professional productivity, with long-horizon tasks such as full-stack development, data analysis, and Office workflows as target use cases. The launch materials do not publish a benchmark table, context-window figure, API specification, or model card. An independent report says Qwen developer Shuai Bai described it as the team's first multimodal model above one trillion parameters, able to process images, video, and documents, but also notes that no benchmark results were available at publication. 5 6
That leaves the product's practical shape narrower than the parameter count suggests. The confirmed access paths today are managed subscription and coding products; a downloadable checkpoint or general-purpose API has not been documented. Alibaba says open weights are coming, but has not supplied a release date, license, checkpoint, or hardware requirement. 1 3
Why it matters
Qwen3.8-Max-Preview extends the recent run of very large Chinese frontier-model launches, arriving only days after Moonshot AI's Kimi K3 drew attention with its own high-end offering. The immediate competitive move is distribution: Alibaba has put a preview in front of eligible developers while promising to widen access later. 2
For developers, the next meaningful checkpoints are the open-weight release, a public model card or API specification, and independent evaluations. Until those appear, Qwen3.8-Max-Preview is an accessible preview with ambitious capability claims, not yet a proven leaderboard result.
相似内容
- 登录后可发表评论。
More from this channel›
- Kimi K3's second week turns a model launch into a compute and policy test
- Qwen-Audio-3.0-TTS splits the launch between speed and voice quality
- OpenAI Presence brings managed voice and chat agents to enterprise workflows
- Google launches Gemini 3.6 Flash, 3.5 Flash-Lite, and Flash Cyber
- Thinking Machines launches Inkling, a 975B open-weight model built for customization
- Kimi K3 brings 2.8T parameters to the open-model frontier — but not yet to your servers
- Meta's Muse Spark 1.1 is live, and speed is the real headline
- GPT-5.6 Sol, Terra, and Luna get a public launch date: July 9