# Alibaba Launches Qwen3.8-Max, a 2.4 Trillion Parameter Open-Weight AI Model > Source: > Published: 2026-08-04 12:23:29.603322+00:00 # Alibaba Launches Qwen3.8-Max, a 2.4 Trillion Parameter Open-Weight AI Model - Qwen3.8-Max has 2.4 trillion total parameters but activates only 95 billion at inference through a sparse mixture-of-experts architecture, with a 1 million token context window [[1]](https://technode.global/2026/08/04/chinas-alibaba-launches-qwen3-8-max-ai-model-with-2-4t-parameters-1m-token-context-window/) - The model ranks fifth in Text Arena, second in Vision Arena, and fourth in Frontend Code Arena with a score of 1,668 — 37 points behind Claude Opus 5 [[2]](https://siliconangle.com/2026/08/03/alibaba-debuts-qwen3-8-max-model-2-4t-parameters/) - Output tokens are priced at roughly 24% of Claude Opus 5's cost, with input tokens at about 40% of Claude Opus 5 pricing [[3]](https://news.cgtn.com/news/2026-08-03/Alibaba-unveils-Qwen3-8-Max-its-most-capable-AI-model-to-date-1Pj3AAwmjPa/p.html) - Alibaba plans to open-source the model weights next week, marking a return to open-weight releases after keeping several recent flagship models proprietary [[1]](https://technode.global/2026/08/04/chinas-alibaba-launches-qwen3-8-max-ai-model-with-2-4t-parameters-1m-token-context-window/) - BABA shares rose more than 4% in pre-market trading Monday following the announcement [[4]](https://www.tradingkey.com/analysis/stocks/us-stocks/262070662-alibaba-baba-stock-forecast-august-3-2026-qwen-ai-earnings-breakout-tradingkey) Alibaba on August 3 released Qwen3.8-Max, the largest model in its Qwen family, featuring 2.4 trillion parameters and a context window of up to 1 million tokens. The model uses a sparse mixture-of-experts architecture that activates only 95 billion parameters per query, reducing computational costs compared to dense models of similar scale [1]. The release positions Alibaba's flagship AI system within striking distance of the top Western frontier models on key benchmarks. Qwen3.8-Max ranks fifth in Text Arena, second in Vision Arena, and fourth in Frontend Code Arena, where its score of 1,668 trails Anthropic's Claude Opus 5 by just 37 points [2]. Alibaba said it plans to release the model's weights for public download next week, resuming an open-source strategy after keeping several recent releases proprietary . [[1]](https://technode.global/2026/08/04/chinas-alibaba-launches-qwen3-8-max-ai-model-with-2-4t-parameters-1m-token-context-window/)The model is now accessible via API on Alibaba Cloud Model Studio. On pricing, Alibaba is undercutting U.S. rivals: input tokens cost roughly 40% of Claude Opus 5's rate, and output tokens about 24%, according to international market pricing [3]. BABA shares jumped more than 4% in pre-market trading Monday on the news . [[4]](https://www.tradingkey.com/analysis/stocks/us-stocks/262070662-alibaba-baba-stock-forecast-august-3-2026-qwen-ai-earnings-breakout-tradingkey)## The Model Qwen3.8-Max is built on the Qwen 3.5 foundation and employs a sparse mixture-of-experts design with a hybrid attention mechanism. Despite its 2.4 trillion parameter count, the model activates only 95 billion parameters at inference time, a design choice that reduces latency and computational expense while maintaining performance [1]. It can generate responses up to 131,000 tokens long . [[2]](https://siliconangle.com/2026/08/03/alibaba-debuts-qwen3-8-max-model-2-4t-parameters/)On multimodal capabilities, Alibaba said the model processes text, images, and video — analyzing 200-plus pages of text or roughly 100 hours of video per request. The company demonstrated examples of the system converting full television series and 100-hour livestreams into searchable, interactive knowledge bases [1]. Alibaba also highlighted autonomous coding abilities: the model completed a 16-day software engineering project without human intervention, producing an open-sourced framework called oh-my-cli. A separate demonstration showed the model performing chip design optimization involving more than 500 steps [2]. ## Competitive Landscape By parameter count, Qwen3.8-Max is the second-largest open-weight model behind Moonshot AI's Kimi K3, which has 2.8 trillion parameters [3]. The model's benchmark performance places it in the top tier globally but still behind Anthropic's Claude family on most leaderboards. The aggressive pricing strategy is notable. At roughly a quarter of Claude Opus 5's output token cost, Alibaba is leveraging infrastructure scale to compete on economics as well as capability [3]. The model is integrated into QwenWork, Alibaba's workplace AI agent platform, which supports applications in legal review, financial research, and architectural modeling . [[1]](https://technode.global/2026/08/04/chinas-alibaba-launches-qwen3-8-max-ai-model-with-2-4t-parameters-1m-token-context-window/)The release comes amid intensifying competition between Chinese and American AI labs. Alibaba's decision to open-source the weights — after a period of keeping flagship models proprietary — signals a strategic shift back toward the community-building approach that made earlier Qwen models widely adopted by developers globally [1]. ## Market Reaction Alibaba's U.S.-listed ADR (BABA) rose more than 4% in pre-market trading on Monday, August 4, following the Sunday announcement. The stock traded in a range of $126.04 to $129.50, up from a previous close of $122.25 [4]. The company's next earnings report is scheduled for August 17. Analysts maintain a consensus Strong Buy rating on BABA, with an average 12-month price target of $189.81 [4]. Alibaba's cloud division, which hosts the model's API, has been a focal point for investors watching the company's AI monetization strategy. ## What's Next Alibaba said it will release the full model weights for public download in the coming week. A smaller variant, Qwen3.8-27B, designed for more hardware-constrained environments, is also planned for open-source release [2]. The open-weight release will allow third-party developers to fine-tune and deploy the model independently of Alibaba Cloud, a move that broadens distribution but also raises familiar questions about safety guardrails on powerful open-weight systems. The model's ability to autonomously execute multi-day coding projects adds urgency to those discussions as capabilities continue to scale. ## Companies mentioned ## Further sources [[1] TechNode Global — China's Alibaba launches Qwen3.8-Max AI model with 2.4T param… ↗](https://technode.global/2026/08/04/chinas-alibaba-launches-qwen3-8-max-ai-model-with-2-4t-parameters-1m-token-context-window/) [[2] SiliconANGLE — Alibaba debuts Qwen3.8-Max model with 2.4T parameters ↗](https://siliconangle.com/2026/08/03/alibaba-debuts-qwen3-8-max-model-2-4t-parameters/) [[3] CGTN — Alibaba unveils Qwen3.8-Max, its most capable AI model to date ↗](https://news.cgtn.com/news/2026-08-03/Alibaba-unveils-Qwen3-8-Max-its-most-capable-AI-model-to-date-1Pj3AAwmjPa/p.html) [[4] TradingKey — Alibaba (BABA) Stock Forecast: Qwen AI, Profitability Recovery, an… ↗](https://www.tradingkey.com/analysis/stocks/us-stocks/262070662-alibaba-baba-stock-forecast-august-3-2026-qwen-ai-earnings-breakout-tradingkey) The stories that matter, in one email. Free — unsubscribe anytime.