Alibaba Qwen3.8-Max Unveiled, Shares Rally

Alibaba Group Holding Limited (BABA) unveiled Qwen3.8-Max on Aug. 3, 2026, a 2.4 trillion-parameter AI model it plans to release as open weights next week. The company framed the launch as a step to convert AI investment into Alibaba Cloud demand, coinciding with a rally in its shares.

Model Specs and Release

Qwen3.8-Max is the newest flagship in Alibaba’s Qwen foundation family and its largest model to date, with 2.4 trillion parameters. The model supports a context window of up to 1 million tokens, enabling it to process and reason over extensive text inputs. Alibaba described it as the most powerful model in the Qwen series and its flagship offering.

The company disclosed that the model serves 95 billion “activated” parameters per request, the operational unit used for inference and pricing. Alibaba plans to release the Max model’s weights publicly next week through Alibaba Cloud’s Model Studio and developer channels. Alongside the Max model, a smaller open-weight variant called Qwen3.8-27B will also be released.

Commercial Strategy and Competition

Qwen3.8-Max is available in a preview tier branded Qwen3.8-Max-Preview on developer programs and platforms including Token Plan, Qoder, QoderWork, and via Qwen AI platform APIs. Alibaba integrated the model into workplace tools under Qwen Office and QwenWork, which merge existing agent tools and will connect to the DingTalk enterprise collaboration app. This integration aims to route office and developer workloads into the Qwen ecosystem and Alibaba Cloud.

Domestic API pricing includes RMB 12 per million input tokens, RMB 36 per million output tokens, and RMB 1.5 per cache-hit request. Promotional preview pricing offers access at roughly 10% of the standard price, with an exclusive night-time rate providing discounts up to 98% for certain developer plans.

Demonstrations highlighted the model’s long-horizon autonomous operation. In one internal test, Qwen3.8-Max worked independently for about 125 hours reconstructing and improving a math research experiment. In a separate 24-hour live contest against 526 human teams, it finished ahead of all but 68. Alibaba also reported a 16-day internal software engineering project executed autonomously by the model.

In the domestic AI race, Moonshot AI’s Kimi K3 model is cited at 2.8 trillion parameters, slightly larger than Qwen3.8-Max. Alibaba positions its model as broadly competitive with leading U.S. models from OpenAI and Anthropic, outperforming them on several coding, multimodal, and engineering benchmarks while trailing on some general-purpose reasoning tests. Comprehensive benchmark tables and methodologies have not been publicly disclosed.

No explicit revenue, profit, or segment-level financial guidance tied directly to Qwen3.8-Max was published in the sampled documents, and no SEC filing referencing the model appeared in the past 72 hours. Investors responded positively to the launch, with Alibaba shares rising on the news.

Alibaba describes making the model widely accessible as part of a broader push to convert years of AI investment into monetizable Alibaba Cloud demand, a strategic framing with direct revenue-growth implications for its cloud business.