The open-weight version, released on Hugging Face and ModelScope as Qwen/Qwen3.8-2.4T-A95B, omits certain cloud-model features such as image input and a non-thinking mode . At 2.4T total parameters, the checkpoint is a multi-node datacenter artifact that requires substantial infrastructure to run
.
Smaller businesses, researchers, and individual developers can use the open weights freely without a separate license . The $50 million threshold is significantly higher than the $20 million annual-sales trigger used by Moonshot AI in its Kimi K3 license, which served as a template for Alibaba's approach
.
Internal use of the model — running it for a company's own operations without offering it as a hosted service or product to others — is explicitly exempted from the commercial license requirement . The restriction targets companies that package and resell the model as an AI assistant or inference service, not enterprises using it behind their own firewalls
.
Alibaba offers Qwen3.8-Max through its hosted API at the following per-million-token rates, confirmed at the August 3, 2026 general availability launch :
International (USD) pricing:
China (CNY) pricing:
The API is OpenAI- and DashScope-compatible, allowing developers to switch model endpoints by changing the base URL and model ID .
Anthropic's Claude Opus 5 is priced at $5 per million input tokens and $25 per million output tokens as of August 2026, holding steady across multiple Opus releases since Opus 4.5 . The pricing differential is stark:
| Rate per 1M tokens | Qwen3.8-Max | Claude Opus 5 | Qwen discount |
|---|---|---|---|
| Input | $2.00 | $5.00 | 60% cheaper |
| Output | $6.00 | $25.00 | 76% cheaper |
Qwen3.8-Max's API undercuts Claude Opus 5 by 60% on input and 76% on output, a pricing advantage that multiple outlets note is central to Alibaba's competitive positioning as it seeks to attract developers and enterprises seeking lower-cost frontier-model access .
The Qwen3.8-Max release represents a broader industry trend away from truly free open-source AI models . By open-sourcing weights for the first time while imposing revenue-based licensing on large commercial users, Alibaba is attempting to capture value from the deployment layer — effectively treating its model weights as a royalty-bearing asset rather than a public good
. The move follows the precedent set by Moonshot's Kimi K3 license and signals that the era of unrestricted open-weight AI from major Chinese labs may be ending for large-scale commercial applications.