Alibaba released Qwen3.8 Max, a 2.4 trillion parameter sparse MoE model, as open weights on August 13, 2026.
What new commercial restrictions did Alibaba introduce with the open-weight release of its Qwen3.8-Max AI model in August 2026, including thAn editorial visualization of Alibaba's Qwen3.8-Max AI model release, illustrating its massive scale and commercial licensing shift.
AI Prompt
Create a landscape editorial hero image for this Studio Global article: What new commercial restrictions did Alibaba introduce with the open-weight release of its Qwen3.8-Max AI model in August 2026, including th. Article summary: Here's a full breakdown of Alibaba's commercial restrictions and pricing for the Qwen3.8-Max open-weight release in August 2026.. Topic tags: general, general web, news, documentation. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts with fake numbers, clickbait thumbnails, icons, and tiny thumbnail layouts. Make it useful as
openai.com
Alibaba made its largest-ever AI model, Qwen3.8-Max, available as open weights on August 13, 2026 — but with strings attached. Instead of a permissive open-source license, the company introduced a tiered commercial framework that requires large-scale users to negotiate a paid license once revenue exceeds $50 million. This marks the first time Alibaba has open-sourced a "Max"-class model, signaling a strategic shift from purely proprietary flagship releases to a dual-track strategy balancing open adoption with enterprise monetization .
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
Alibaba released Qwen3.8 Max, a 2.4 trillion parameter sparse MoE model, as open weights on August 13, 2026.
What new commercial restrictions did Alibaba introduce with the open-weight release of its Qwen3.8-Max AI model in August 2026, including thAn editorial visualization of Alibaba's Qwen3.8-Max AI model release, illustrating its massive scale and commercial licensing shift.
AI Prompt
Create a landscape editorial hero image for this Studio Global article: What new commercial restrictions did Alibaba introduce with the open-weight release of its Qwen3.8-Max AI model in August 2026, including th. Article summary: Here's a full breakdown of Alibaba's commercial restrictions and pricing for the Qwen3.8-Max open-weight release in August 2026.. Topic tags: general, general web, news, documentation. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts with fake numbers, clickbait thumbnails, icons, and tiny thumbnail layouts. Make it useful as
openai.com
Alibaba made its largest-ever AI model, Qwen3.8-Max, available as open weights on August 13, 2026 — but with strings attached. Instead of a permissive open-source license, the company introduced a tiered commercial framework that requires large-scale users to negotiate a paid license once revenue exceeds $50 million. This marks the first time Alibaba has open-sourced a "Max"-class model, signaling a strategic shift from purely proprietary flagship releases to a dual-track strategy balancing open adoption with enterprise monetization .
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
Qwen3.8-Max is a 2.4 trillion-parameter model using a sparse Mixture-of-Experts (MoE) architecture, with approximately 95 billion parameters activated per token . It supports a 1-million-token context window with up to 131,000 output tokens and handles text, image, and video inputs natively . Alibaba has described the model's performance as "second only to Claude Fable 5," though independent benchmarks remain limited at launch .
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 .
The $50 Million Revenue Threshold
The custom Qwen3.8-Max License (not Apache 2.0) introduces a two-tier commercial restriction :
$50 million trigger: Any business operating as a "Model as a Service" (MaaS) or "AI Work Assistant" whose aggregate revenue — including affiliates — exceeds $50 million in any consecutive 12-month period must obtain a separate paid commercial license from Alibaba before using the model or its derivatives for commercial purposes .
Attribution rider: Products or services with more than 100 million monthly active users or $20 million in monthly revenue must prominently display the model name on their user interface .
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 Exemption
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 .
API Pricing
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:
Input: $2 per million tokens
Output: $6 per million tokens
Cached input: $0.25 per million tokens
China (CNY) pricing:
Input: 12 yuan per million tokens
Output: 36 yuan per million tokens
The API is OpenAI- and DashScope-compatible, allowing developers to switch model endpoints by changing the base URL and model ID .
Comparison to Claude Opus 5
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 .
Strategic Implications
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.
developersdigest.techQwen 3.8 Max Ships: 2.4T MoE, 1M Context, $2/$6 per MTok, Open ...
Qwen3.8-Max is a 2.4 trillion-parameter model using a sparse Mixture-of-Experts (MoE) architecture, with approximately 95 billion parameters activated per token . It supports a 1-million-token context window with up to 131,000 output tokens and handles text, image, and video inputs natively . Alibaba has described the model's performance as "second only to Claude Fable 5," though independent benchmarks remain limited at launch .
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 .
The $50 Million Revenue Threshold
The custom Qwen3.8-Max License (not Apache 2.0) introduces a two-tier commercial restriction :
$50 million trigger: Any business operating as a "Model as a Service" (MaaS) or "AI Work Assistant" whose aggregate revenue — including affiliates — exceeds $50 million in any consecutive 12-month period must obtain a separate paid commercial license from Alibaba before using the model or its derivatives for commercial purposes .
Attribution rider: Products or services with more than 100 million monthly active users or $20 million in monthly revenue must prominently display the model name on their user interface .
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 Exemption
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 .
API Pricing
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:
Input: $2 per million tokens
Output: $6 per million tokens
Cached input: $0.25 per million tokens
China (CNY) pricing:
Input: 12 yuan per million tokens
Output: 36 yuan per million tokens
The API is OpenAI- and DashScope-compatible, allowing developers to switch model endpoints by changing the base URL and model ID .
Comparison to Claude Opus 5
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 .
Strategic Implications
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.
developersdigest.techQwen 3.8 Max Ships: 2.4T MoE, 1M Context, $2/$6 per MTok, Open ...
