Alibaba launches Qwen4 Max AI model with up to 67% lower task cost
Alibaba's new Qwen4 Max AI model cuts per‑task costs by two‑thirds while delivering higher performance, prompting a modest share rise but raising questions about revenue impact.
Alibaba's new Qwen4 Max AI model cuts per‑task costs by two‑thirds while delivering higher performance, prompting a modest share rise but raising questions about revenue impact.
Alibaba Group announced the launch of its latest large language model, Qwen4 Max, on Monday, highlighting a dramatic reduction in operating costs while delivering a performance boost. The new model is priced at roughly a third of the cost per token of its predecessor, Qwen3.8 Max, and is being offered through Alibaba Cloud’s model‑as‑a‑service (MaaS) platform. The announcement comes as the company’s Hong Kong‑listed shares have fallen nearly 39% year‑to‑date, prompting analysts to assess whether the cheaper model can offset lower per‑task revenue with higher usage.
Independent benchmarker Artificial Analysis measured Qwen4 Max at 43 on its Intelligence Index, nine points above Qwen3.8 Max, and placed it 14th on the Arena Agent leaderboard – the highest rank among Chinese models. The cost reduction stems from multiple efficiency gains: per‑token pricing is 25% lower for inputs and 70% lower for cached‑input tokens, while the model uses 39% fewer output tokens and 32% fewer input tokens per task. As a result, the cost per task fell from $10.02 to $3.27 on the Artificial Analysis index (a 67% drop) and from $4.43 to $1.67 on Alibaba’s internal Work360 benchmark (a 62% drop).
Alibaba expects its AI offerings to become the largest source of cloud revenue, with MaaS and related services already generating RMB 16 billion in annual recurring revenue as of August. However, the lower price means each task generates less revenue, so total earnings will depend on whether usage scales sufficiently to compensate. Early internal tests showed a 10.42% cost saving on data‑labeling scheduling, suggesting the leaner design could protect margins, but the market impact remains uncertain until usage data for the new pricing model are released.
Following the launch, Alibaba’s Hong Kong‑listed shares edged up 1.05% to HK$91.50 during the lunch break, a modest move compared with the 6.15% rise seen when Qwen3.8 Max debuted in August. The stock still trades about 19% below the HK$112.70 price set for the company’s $10 billion share sale earlier this year. Alibaba will report its September‑quarter results on November 19, which do not yet reflect Qwen4 Max activity. The first earnings period to include the new model will be the December quarter, providing the first clear signal of whether the lower‑priced offering can drive sufficient usage growth.
Investors will be looking for usage metrics in Alibaba’s upcoming December‑quarter earnings, particularly the volume of API calls to Qwen4 Max and any changes in average revenue per task. Comparisons with rival models, such as Tencent’s Hy4, which costs less per task but scores lower, will also shape expectations for Alibaba’s AI‑cloud strategy.
Discussion 0