EL7.AI
StocksMedium•16 November 2026•
5 min read

Alibaba cuts average cost of Qwen4-Max tasks by about 60% on its internal benchmark

Key Facts

1Alibaba Cloud announced the launch of Qwen4-Max on November 16, 2026.
2Qwen4-Max cut average task costs by about 60% on the Work360 benchmark, according to the company.
3The model has 2.4 trillion parameters and a context window of up to one million tokens.
4Revenue from cloud services and AI computing reached 48,437 million yuan in the quarter ended June 30, 2026.
5Alibaba will announce results for the quarter ended September 30, 2026, before the U.S. market opens on November 19, 2026.

Alibaba Cloud announced the launch of Qwen4-Max on November 16, 2026, describing it as the most capable model in the Qwen family to date. The company said the model cut average task costs by about 60% compared with its predecessor on the Work360 benchmark, linking model performance with the cost of using it. The benchmark covers 721 specialized fields, while the model used 35% fewer tokens in high-effort reasoning mode. These figures come from a test conducted by the company itself, so they reflect the benchmark results it announced rather than an independent market measurement. Cost reduction is a central part of the model’s launch for developers and companies using AI services.

Qwen4-Max has a total of 2.4 trillion parameters and supports native image processing, with a context window of up to one million tokens. According to the specifications announced by Alibaba Cloud, the long context window lets the model handle large inputs within a single context. The company said the model is built on a mixture-of-experts architecture and activates 71 billion parameters per token. That is 25% fewer activated parameters than in Qwen3.8-Max, while both models have the same total size of 2.4 trillion parameters. The company presents this architecture as part of its explanation for the lower token use and cost in the Work360 test.

TechNode reported that average task costs on the Work360 test fell from 26.58 yuan to 11.16 yuan, a decline of about 58%. That result is close to the roughly 60% cost reduction announced by Alibaba Cloud, although the figures presented by the two sources differ. TechNode also reported that public access through QwenCloud followed a limited preview that began on November 11, 2026. The limited test thus preceded the official announcement by five days, before the model moved to public access, according to the report. These figures give users a direct comparison of costs before and after the model’s introduction on the announced benchmark.

In a company presentation on the model’s use in manufacturing, the time needed to program a workpiece manually fell by 95%, from about 475 minutes to 24 minutes. The example presents a specific industrial use case among the materials Alibaba Cloud published about the model. In another test of cloud resource scheduling, the company said the average cost of calls to the data annotation model fell by 9%. It added that the speed of rescheduling offline CPUs rose to more than 14 times its previous level. These results relate to the cloud resource scheduling test, while the 9% reduction applies specifically to the cost of calls reported in that test.

Alibaba Cloud made the model available through QwenCloud via its token plan and API, and also through QwenWork and QwenStudio. The company set the price for input tokens with an implicit cache hit at $0.075 per million tokens. This price links the model launch to the paid usage services the company offers through its platforms. The pricing announcement presented here does not include other usage-cost figures, so the price comparison is limited to this type of input. Alibaba said smaller open-weight models from the Qwen4 series would be released to developers globally later, without specifying a date.

The model launch comes as Alibaba expands its cloud and AI business. Revenue from cloud services and AI computing reached 48,437 million yuan in the quarter ended June 30, 2026. The group reported that revenue rose 45% from the same period in 2025, attributing the growth mainly to public cloud growth and increasing adoption of AI products. In the same quarter, free cash flow recorded an outflow of 44,670 million yuan, compared with an outflow of 18,815 million yuan in the corresponding period of 2025. The company attributed the decline mainly to increased spending on cloud infrastructure. The figures show revenue growth alongside higher spending tied to infrastructure expansion.

Alibaba said it plans to invest at least 380 billion yuan over three years in cloud computing and AI infrastructure. It also set a target of exceeding $100 billion in annual AI and cloud revenue over the next five years. In the quarter ended in June, capital expenditure rose 75% to 67.7 billion yuan, with the company attributing the spending to infrastructure investment to meet demand. The model launch forms part of a broader expansion in spending and business, according to the targets and figures announced by the group. The next quarter’s results are a near-term milestone for tracking the company’s financial performance after these figures.

Alibaba scheduled the announcement of results for the quarter ended September 30, 2026, before the U.S. market opens on November 19, with a call to discuss the results on the same day. In the stock market, EL7 data recorded BABA’s close at $94.38 on November 13, the latest listed close before the news date. The same data recorded a close of $105.38 on November 9, before the stock fell to $94.38 on November 13. These prices were recorded before the model launch announcement on November 16 and therefore do not represent a market reaction to the announcement. The November 19 results are the next announced date for tracking the group’s financial figures.

Open on chart