Buy
Market
đŸ”„
Prediction Market

Alibaba's Qwen4 Max Costs Two-Thirds Less Than Predecessor

Alibaba launched Qwen4 Max at roughly a third of its predecessor's cost, as shares fell 38.59% in Hong Kong since January 1.

16/11/2026 11:2615 min read

Alibaba introduced Qwen4 Max on Monday, with a cost per task roughly one-third that of its previous version. The cheaper model comes as Alibaba shares traded in Hong Kong have fallen 38.59% since the start of the year.

For those holding Alibaba stock, the reduced price involves a compromise. Each task now generates less revenue for Alibaba, meaning total income declines unless usage rises sufficiently to make up the difference.

How Much Smarter and Cheaper Is Qwen4 Max?

Independent AI benchmarking firm Artificial Analysis gave Qwen4 Max a score of 43 on its Intelligence Index. That represents a nine-point improvement over Qwen3.8 Max.

On Arena's Agent leaderboard, which is based on real-world user tasks, Qwen4 Max sits in 14th place and ranks first among Chinese models.

Qwen4-Max by @Alibaba_Qwen is the #1 Chinese model on the leaderboard making Alibaba now the #1 Chinese lab and #5 lab overall, just behind Anthropic, Google, OpenAI and Meta.

It debuts overall at #14 on Agent Arena with +7.5% net improvement across 14.7k real-world agentic
 https://t.co/HLnmxkqvUY https://t.co/wghkKeXcsP

— Arena.ai (@arena) November 16, 2026

The cost difference with Qwen3.8 Max is larger than the score difference. Alibaba reduced per-token prices by 25% and cut cached-input prices by 70%.

“Qwen4 Max uses ~105k output tokens per task, 39% fewer than Qwen3.8 Max (~171k), and 32% fewer input tokens to run the Index (9.9B vs 14.5B). Input makes up ~85-90% of the cost of running the Index for both models since agentic evaluations re-send context on every turn, so the 70% lower cache hit price ($0.075 vs $0.25 per 1M tokens) and 25% lower input price ($1.50 vs $2.00) for Qwen4 Max reduce Cost per Task,” Artificial Analysis explained.

Combined with reduced token usage, the total savings differ across tests.

  • On the Artificial Analysis index, the cost per task declined 67%, moving from $10.02 to $3.27.
  • On Work360, Alibaba's own test of industry tasks, the average cost per task dropped 62%, from $4.43 to $1.67.
  • Alibaba reports that token usage fell by 24% to 35% on Work360, depending on the reasoning configuration. On Terminal-Bench 4.0, a test of command-line tasks, the decline was about 48% to 49%.

These cost figures matter to shareholders because Alibaba expects AI model sales to become its largest cloud revenue source.

Can a Cheaper Qwen Lift Alibaba Cloud Revenue?

Alibaba offers Qwen through Alibaba Cloud as model-as-a-service (MaaS), with per-token pricing for API customers. Its model and application services, including MaaS, had passed RMB 16 billion in annual recurring revenue by August.

Qwen4 Max activates 25% fewer parameters per token compared with Qwen3.8 Max. Alibaba states that the new design improves computing efficiency.

The company has also deployed the model on its own cloud. In one instance, the model reduced the cost of scheduling model calls for data labeling by an average of 10.42%.

These efficiency improvements could help cushion the margin impact of lower prices. Still, it remains uncertain whether usage will increase enough to offset the price reductions.

Hong Kong trading on Monday gave investors their first opportunity to price the launch.

What Does the Launch Mean for Alibaba Stock?

Alibaba shares in Hong Kong rose 1.05% to HK$91.50 by Monday's lunch break, up from Friday's close of HK$90.55.

By comparison, the shares climbed 6.15% by the same break on the day Qwen3.8 Max was launched in August. The shares currently trade nearly 19% below HK$112.70, the price set for Alibaba's $10 billion share sale in August.

Alibaba reports September-quarter results on Thursday, November 19. That quarter ended before Qwen4 Max was released, so the new model does not appear in those figures. The first results to include it will cover the December quarter.

However, the period includes Qwen3.8 Max's first two months on the market. Any update on MaaS would show how model demand tracked before the most recent price reduction.

Until that point, the case for the stock depends on two unproven factors. Usage must grow faster than per-task revenue declined, and the more efficient design must safeguard margins.

Tencent's Hy4 costs even less, at $1.03 per task on the Artificial Analysis index, but scores eight points lower.

Share to

Disclaimer: this article comes from third-party media and is provided for reference only. It does not constitute investment advice. Crypto and other financial products carry significant price volatility risk, so please make your own decisions carefully.

Related articles