What happened
Alibaba Cloud’s Model Studio pricing documentation lists separate input- and output-token prices for Qwen4-Max and Qwen3.8-Max. In Singapore, Qwen4-Max is priced at $1.50 per million input tokens and $4.50 per million output tokens. The same table gives Qwen3.8-Max rates of $2 per million input tokens and $6 per million output tokens. That makes each listed Qwen4-Max rate 25% lower in this region.
The table also shows lower Qwen4-Max rates in Beijing and Global regions. In both, Qwen4-Max is listed at $1.413 per million input tokens and $4.24 per million output tokens, compared with $1.65 and $4.951 for Qwen3.8-Max. The price difference is not identical in percentage terms to Singapore because the listed rates differ by region and amount. These are per-million-token rates, so the bill for a particular application depends on its input and output token volumes.
The comparison puts a concrete unit-price difference between the two named models in front of teams using Model Studio. It does not by itself describe how either model performs on a coding, reasoning, extraction, or agent workflow. The practical question is therefore not only which row has the lower price, but whether Qwen4-Max can satisfy the same acceptance criteria with a comparable number of calls and review steps.
For cost modelling, builders can apply the listed input rate to prompt and context volume and the output rate to generated tokens, separately for the region they expect to use. Long responses make output pricing particularly salient, while applications that send large contexts should pay close attention to input rates. Retries and verification can also change the effective spend, so the unit rates are a starting point rather than a complete task-cost calculation.
What the archive adds
New Runtime noted on November 11, 2026, that Qwen4-Max Preview was listed in QwenCloud’s Token Plan Individual documentation for a two-week preview ending November 25, 2026 (earlier coverage). This pricing comparison is a new fact on a known Qwen4-Max story: it adds regional per-token rates alongside a comparison with Qwen3.8-Max.
Trend signal
This supports the view that AI API pricing is moving beyond a single flat per-token rate toward differentiated terms and options (compute-backed AI contracts). Here, the concrete signal is a regional price comparison between two model generations, rather than a new capacity-linked tariff. For buyers, the useful next step is to compare cost against measured task outcomes, not to treat a lower token price as proof of lower cost per successful task.
Why it matters
Published unit prices give teams a concrete way to identify candidates for evaluation before they change a production route. At the listed Singapore rates, Qwen4-Max is 25% lower on both input and output tokens. That is a clear difference in the price table, but it does not guarantee the same percentage reduction in a monthly bill: actual savings depend on traffic volume, input/output mix, retries, and the work required to review outputs.
For a team already using Alibaba Cloud, the comparison is a prompt to run a controlled evaluation rather than assume a cheaper or newer model is an automatic replacement. Use representative requests, define acceptance criteria before testing, and count unsuccessful attempts and verification work alongside token spend. If Qwen4-Max clears the same bar for the same workload, the lower listed rates could reduce variable API costs. If it requires extra calls or human correction, a lower price per token may not translate into lower operating cost.
The distinction matters for applications with variable or growing usage. A procurement estimate built only from a headline model price can obscure how prompts, long context, and generated output contribute differently. Separating input and output volumes and applying the relevant regional rates makes the comparison more useful—and easier to revisit when traffic or model choices change.
Limits and uncertainty
Alibaba Cloud’s pricing table gives regional token rates, but these figures do not establish comparative model quality, latency, uptime, or the price of completing a specific task. The listed Singapore rates are 25% lower for Qwen4-Max on each token category, but that is not evidence that a workload will cost 25% less overall. Token volumes, input/output mix, retries, verification, and applicable account or service terms can affect the bill. Teams should confirm current rates for their account and region and benchmark both models on their own tasks before switching production traffic.
Reader takeaway
Qwen4-Max is listed below Qwen3.8-Max per token in Singapore, Beijing, and Global regions, with a 25% lower Singapore rate for both input and output. Treat that gap as a reason to benchmark, not as a measured savings result. Compare task quality, token use, retries, and review effort on your own workload before rerouting calls.