By Jonathan West · Updated Nov 17, 2026

Qwen4-Max Pricing per Token and per Task

Its $3.27 cost per task on Artificial Analysis Intelligence Index v5.0.1 is lower than the two listed alternatives, but your token use determines your bill.

By Jonathan West · Updated Nov 17, 2026

Qwen4-Max costs $1.50 per million input tokens and $4.50 per million output tokens on the international Alibaba Cloud Model Studio endpoint. You can also run it through OpenRouter or Vercel AI Gateway at the same listed input and output rates.

On the Artificial Analysis Intelligence Index v5.0.1 Qwen4 Max page, checked November 16, 2026, its reported cost is $3.27 per task. That figure comes from Artificial Analysis's workload and live cache-hit measurements. It is not a fixed charge for each request you send.


Qwen4-Max Pricing per Token and per Task

Qwen4-Max has a $1.50 input rate and a $4.50 output rate per million tokens on Alibaba Cloud Model Studio's international Singapore endpoint. The published tier covers requests with 0 to 1 million input tokens. Alibaba lists the same rates for thinking and non-thinking modes.

Price measureQwen4-Max figureWhat the figure covers
Input$1.50 per million tokensInternational list price for tokens sent to the model
Output$4.50 per million tokensInternational list price for tokens generated by the model
Intelligence Index task$3.27Artificial Analysis's measured workload and live cache-hit rates

The per-task figure answers a different question from the token rates. Artificial Analysis calculates a weighted average cost for its Intelligence Index tasks using provider token prices and live cache-hit rates. Your cost changes with the amount of input and output your requests use. Use the published token rates for a workload estimate. Use the index figure to compare models on Artificial Analysis's workload.

Alibaba's production listing gives Qwen4-Max a 1,000,000-token context length. It lists a maximum non-thinking output of 131,072 tokens and a maximum combined thinking and answer output of 262,144. If you budget a request that uses thinking, account for those tokens within the combined limit.


Qwen4-Max Pricing by Alibaba Region

Alibaba Cloud Model Studio lists a lower Qwen4-Max token price on five regional rows than on its international Singapore row. Its production pricing page, updated November 16, 2026, gives the following rates for the 0 to 1 million input-token tier.

EndpointInput per million tokensOutput per million tokens
International (Singapore)$1.50$4.50
China (Beijing)$1.413$4.24
Hong Kong (China), Global$1.413$4.24
Germany (Frankfurt), Global$1.413$4.24
United States (Virginia), Global$1.413$4.24
Japan (Tokyo), Global$1.413$4.24

The Singapore and Beijing endpoints use consumption-based rate-limit tiers. The published tiers run from 5 million to 20 million tokens per minute. Virginia, Frankfurt, Hong Kong and Tokyo each list 30,000 requests per minute and 5 million tokens per minute. Compare those limits with your expected request and token rates.

The international row includes a free quota of 1 million tokens. Alibaba says it lasts 90 days from Model Studio activation, model release or application approval, whichever occurs later. Keep that quota separate from a recurring monthly estimate. Alibaba lists explicit and implicit caching as supported, but does not list batch inference, fine-tuning or model deployment as supported for Qwen4-Max.


Cost per Intelligence Index Task Against Qwen3.8 Max and Kimi K3.1 Pro

Qwen4-Max has the lowest reported cost per task of these three models on Artificial Analysis Intelligence Index v5.0.1. All three model pages were checked on November 16, 2026. Comparing scores from a different index version would change the basis of the decision.

ModelIndex scoreCost per taskOutput tokens in indexInput / output per million tokens
Qwen4-Max43$3.27127 million$1.50 / $4.50
Qwen3.8 Max (0902)34$10.02150 million$2.00 / $6.00
Kimi K3.1 Pro46$3.68113 million$2.00 / $10.00

In our internal cost modeling at Layer3Labs, we calculated the differences from those published figures: ($10.02 − $3.27) ÷ $10.02 gives Qwen4-Max a 67.4% lower cost per task than Qwen3.8 Max. The same calculation against Kimi K3.1 Pro gives an 11.1% difference. These calculations use Artificial Analysis's results and do not measure a separate workload.

The $10.02 comparison reflects more than list prices. Qwen4-Max's input and output list rates are each 25% below Qwen3.8 Max's, while its 127 million index output tokens are 15.3% below Qwen3.8 Max's 150 million. Artificial Analysis also uses live cache-hit rates in its per-task cost. See Qwen3.8 Max pricing for the previous model's rates.

Kimi K3.1 Pro is the counterexample to choosing solely by cost. It scores 46 against Qwen4-Max's 43 and costs $0.41 more per measured task. If those three additional index points matter more to you than $0.41 on this workload, compare Kimi K3.1 Pro pricing before choosing. Artificial Analysis shows Qwen4-Max's time per task as unknown, so the table cannot support a speed verdict.


Where to Run Qwen4-Max

Alibaba Cloud Model Studio, OpenRouter and Vercel AI Gateway each list a way to run Qwen4-Max. Alibaba's production pricing page was retrieved on November 16, 2026. Compare the endpoint and cache terms before treating the matching $1.50 and $4.50 base rates as an identical bill. For what Alibaba has released and announced in the Qwen4 series, see Qwen 4 explained.

Access routeListed model IDInput / output per million tokensPublished detail
Alibaba Cloud Model Studioqwen4-max$1.50 / $4.50 internationallyRegional rates vary. An international free quota is listed
OpenRouterqwen/qwen4-max$1.50 / $4.50Lists Alibaba Cloud International as its sole provider
Vercel AI Gatewayalibaba/qwen4-max$1.50 / $4.50Lists Alibaba Cloud as provider

OpenRouter lists a $0.075-per-million cache-read price and a $1.875-per-million five-minute cache-create price. Its displayed weighted average input price is $0.1656 per million tokens because the listing reports a 93.7% cache-hit rate. That average describes traffic on OpenRouter. Do not substitute it for the $1.50 list rate when you have no cache-hit estimate for your requests.

Vercel AI Gateway lists cache reads at $0.08 and writes at $1.88 per million tokens. It also lists $5 in AI Gateway credits every 30 days for free users. Vercel displays a 384K maximum output, while Alibaba's production page gives a 262,144-token combined thinking-and-answer maximum. Use the documented limit for the route and request mode you intend to use rather than treating those two figures as interchangeable. The AI model pricing directory offers more models to compare.


Qwen4-Max Pricing for a Monthly Workload

A modeled month of 100 million input tokens and 20 million output tokens costs $240 for Qwen4-Max at Alibaba's international list rates. This is our internal cost calculation, using published token rates with no caching, free quota, regional discount or reasoning-token adjustment.

ModelInput calculationOutput calculationModeled monthly total
Qwen4-Max100 × $1.50 = $15020 × $4.50 = $90$240
Qwen3.8 Max100 × $2.00 = $20020 × $6.00 = $120$320
Kimi K3.1 Pro100 × $2.00 = $20020 × $10.00 = $200$400

The monthly table holds the input and output volumes constant across models. Artificial Analysis's per-task comparison does not: the models produced different amounts of output on its index. Use the monthly table when you can estimate your token volumes. Use the index table when you want to compare the reported cost of completing Artificial Analysis's fixed tasks.

The difference can narrow if a request uses substantially more Qwen4-Max tokens than the index workload. At the same token mix, Qwen4-Max would need about 3.06 times its Artificial Analysis token use per task to cost as much as Qwen3.8 Max at Qwen3.8 Max's measured use. That break-even calculation is a reason to record input and output tokens before projecting an agent's monthly bill.


Qwen4-Max Pricing Fit and Change Conditions

Qwen4-Max is the lower-cost choice on the reported Intelligence Index task among these three models, but that recommendation does not fit every requirement. Artificial Analysis describes Qwen4-Max as proprietary. If you need open weights or deployment outside the listed hosting routes, choose a model that publishes the access you require instead.

If you require a specific data-residency arrangement, do not infer it from a regional price row. The listed endpoints identify locations and rates, but the published pricing pages do not state every data-handling term. Review the Alibaba Cloud Model Studio Qwen4-Max documentation for the route you are considering.

Our cost recommendation would change if Alibaba changed its list prices or Artificial Analysis remeasured the models under another index version. A workload using far more Qwen4-Max reasoning tokens than the measured index tasks could also reverse the per-task result. For a purchase decision, calculate your expected input and output totals at the published Qwen4-Max rates, then compare that estimate with the hosting route you plan to use.

Frequently Asked Questions

  • No. Alibaba's preview documentation, updated November 16, 2026, says qwen4-max-preview remains available through Token Plan until 24:00 on November 25, 2026 (UTC+8). Alibaba recommends the available production model, Qwen4-Max. The preview will be retired after its preview period ends.
  • Alibaba publishes per-token prices for the production Qwen4-Max model. The separate qwen4-max-preview model is available through Token Plan only and has no pay-as-you-go option, according to Alibaba's preview page.
  • Yes. Alibaba's international Qwen4-Max listing includes a 1-million-token free quota valid for 90 days from Model Studio activation, model release or application approval, whichever is later. Vercel separately lists $5 in AI Gateway credits every 30 days for free users.

Estimate Your Qwen4-Max Workload

Map expected input, output and cache use before choosing a Qwen4-Max hosting route. Book a consultation with Layer3Labs to plan the cost model.

Book a Consultation