Technology

Alibaba Opens Qwen4-Max to Pay-as-You-Go API Users

The production model has US (Virginia)/Global list rates of $1.413 per million input tokens and $4.24 per million output tokens. Its separate, subscriber-only preview is scheduled to retire after November 25.

Alibaba released the production Qwen4-Max model on November 16 with a priced, pay-as-you-go API, adding a purchasing option to the subscriber-only preview introduced five days earlier. For developers and businesses evaluating an AI model, the immediate change is access to the production model without the preview’s Token Plan requirement. Alibaba’s release log lists the two as separate models.

In Alibaba Cloud’s Model Studio price table, Qwen4-Max lists at $1.413 per million input tokens and $4.24 per million output tokens for US (Virginia)/Global. For Singapore/International, its rates are $1.50 and $4.50, compared with $2 and $6 for Qwen3.8-Max. That makes Qwen4-Max’s listed input and output rates 25% lower than its predecessor’s in the Singapore/International table. The comparison is between unit prices, not a prediction of any customer’s total bill.

Capabilities and service limits

Alibaba’s production-model documentation lists text, image and video input, text output, a one-million-token context window and tool calling. It also lists OpenAI- and Anthropic-compatible API formats. Those specifications give teams concrete features to check against an application’s requirements alongside the regional prices.

The same documentation lists Singapore and US (Virginia) among available regions, while marking batch inference, fine-tuning and model deployment unsupported. Buyers that need those functions should account for that distinction when evaluating the production API.

Safety and task-cost claims

In its launch announcement, Alibaba says it filtered offensive cyberattack training material while retaining defensive material and carried out work addressing chemical, biological, radiological and nuclear risks. These are Alibaba’s descriptions of its safeguards, not independently measured safety results.

Alibaba also reports approximately 60% lower average cost per task than Qwen3.8-Max on Work360, its proprietary evaluation spanning 721 specialized fields. That company-reported task benchmark is distinct from the 25% Singapore/International list-rate difference.

The preview documentation says the earlier model remains Token Plan-only and is scheduled to retire after 24:00 November 25, 2026, UTC+8. Teams moving from that preview can now assess the production model under its separate model ID, API terms and applicable regional rates.

Read the source material

Sources & references

How we work with sources ↗

Continue reading

More in Technology

Search the publication

Search the articles published on this site.