DevOps / SRE / Platform · 03.08.2026, 19:33 UTC
Alibaba’s AI coded for 16 days straight and every commit is on GitHub
| Schweregrad | info |
|---|---|
| Kategorie | DevOps / SRE / Platform |
| Quelle | The New Stack ↗ |
| Veröffentlicht | 03.08.2026 UTC |
Sicherheitsmeldung mit Schweregrad noch nicht bewertet. Technische Details im Tab „Originaltext“; empfohlene Schritte in der Checkliste.
Alibaba has launched Qwen3.8-Max, a multimodal model with 2.4 trillion parameters and built for complicated tasks that may take several days to complete. Now available through QwenCloud and Alibaba Cloud Model Studio, it’s priced at $2 per million input tokens and $6 per million output tokens.
Qwen3.8-Max builds on the Qwen3.5 architecture and uses a sparse mixture-of-experts design with hybrid attention. Although the model contains 2.4 trillion parameters in total, it activates approximately 95 billion for each token.
The mixture-of-experts architecture draws on the parts of the model needed for each task — instead of running all 2.4 trillion parameters every time. According to the company, the model weights will be published on Hugging Face and ModelScope next week, making it the first Qwen-Max model released with downloadable weights.
The model’s mixture-of-experts architecture draws on the parts of the model needed for each task, which still leaves developers with an enormous model to host.
How the harness works
Even with only 95 billion parameters active at once, the full weights must be stored and distributed across multiple high-memory GPU nodes. In practice, that puts self-hosting beyond the reach of most developers and makes Qwen3.8-Max a more realistic option for large organizations and inference providers with the infrastructure to run it.
quot
Qwen3.8-Max is a realistic option for large organizations and inference providers with the infrastructure to run it.
This brings Alibaba close to Moonshot AI’s new Kimi K3, which has 2.8 trillion parameters. …