Alibaba Model Releases, Tracked

Alibaba has released Qwen3.8-Flash-Next, an open-weight 125B multimodal MoE model activating only 6B parameters per token under the Qwen Community 1.0 license, previewing the next-generation Qwen4 architecture.

Alibaba released the open-weight checkpoint for Qwen 3.8-Max on August 12-13, 2026 under a custom license. Designated Qwen3.8-2.4T-A95B, the weights cover the full 2.4-trillion-parameter MoE architecture with 95B active parameters — but the public checkpoint is text-only, without the multimodal and 1M-context capabilities of the API version.

Alibaba's Tongyi Lab released Qwen 3.8-27B on August 14, 2026 — a 27.8-billion parameter dense vision-language model under Apache 2.0. Supporting text, image, and video inputs with a native 262K context window, it runs in roughly 16-17 GB of VRAM when quantized to 4-bit, making it viable on a single RTX 4090.

Alibaba launched Qwen 3.8 Max on August 3, 2026 — a 2.4-trillion-parameter sparse MoE model with 95B active parameters, a 1-million-token context window, and native multimodal input (text, image, video, documents). It is available now via Alibaba Cloud Model Studio and QwenWork, with open weights promised for the week of August 10. Alibaba positions it as second only to Claude Fable 5 on frontier benchmarks, and it holds the top spot among Chinese models on the LMSYS Chatbot Arena.

Alibaba released Qwen 3.7 Flash on July 27, a cost-optimised multimodal model priced at $0.03 input / $0.13 output per million tokens with a 1M context window. It accepts text, image, and video input and is designed for high-volume agent loops, browser automation, and support tooling. It is the flash-tier counterpart to the Qwen 3.7 Max, Alibaba's flagship agentic model released in May.