Alibaba Unveils Qwen3.8-Max, a 2.4 Trillion-Parameter Model Marking a Return to Open-Weight Flagship Releases
Alibaba made its new flagship Qwen3.8-Max, a 2.4 trillion-parameter multimodal model, broadly accessible ahead of an open-weights release, sending its Hong Kong-listed shares up 7 percent.
Overview
Alibaba Group Holding has made its next-generation flagship artificial intelligence model, Qwen3.8-Max, widely accessible to global users ahead of an open-weights release, according to South China Morning Post. The move marks Alibaba’s return to open-sourcing its top-tier AI models after keeping several recent flagship releases proprietary earlier this year, SCMP reports.
What We Know
Qwen3.8-Max is a 2.4 trillion-parameter mixture-of-experts model that accepts text, image, and video input and returns text, according to MarkTechPost. It supports a context window of up to 1 million tokens, meaning it can process roughly 750,000 words in a single query, per SCMP. While the headline parameter count is 2.4 trillion, the mixture-of-experts design activates only about 95 billion parameters for any given request, a way of keeping a very large model cheaper to run, according to The Next Web. At 2.4 trillion parameters, Qwen3.8-Max sits just below Moonshot AI’s Kimi K3, which carries 2.8 trillion parameters and which The Machine Herald previously reported rattled Chinese tech and chip stocks on its release.
The model is available to global developers via Alibaba Cloud’s Model Studio APIs, as well as through QwenWork, the company’s workplace AI agent platform, which entered public beta the same day through web and desktop apps, according to SCMP. SCMP describes QwenWork as an all-in-one productivity platform targeting Tencent’s WorkBuddy, Moonshot AI’s Kimi Work, and overseas rivals including Claude Cowork and ChatGPT Work. The hosted API is deployable immediately and maintains OpenAI- and DashScope-compatible interfaces, requiring only a base-URL and model-ID change, MarkTechPost reports. Open weights for Qwen3.8-Max, along with a smaller Qwen3.8-27B checkpoint positioned for on-premise GPU hardware, are due to follow, per MarkTechPost and SCMP.
On pricing, MarkTechPost reports the hosted API is listed at $2.00 per 1 million input tokens and $6.00 per 1 million output tokens, with cached input priced separately at $0.25 per 1 million tokens. On benchmarks, MarkTechPost reports Qwen3.8-Max scored 86.6 on Terminal-Bench 2.1, ahead of Claude Opus 4.8’s 84.6 but behind GPT-5.6 Sol’s 88.8, with the largest gains over its predecessor coming in agentic and multimodal tasks rather than raw reasoning. Separately, on the crowdsourced Arena.AI leaderboard, Qwen3.8-Max ranks as the highest-scoring Chinese text model and second in the world on the platform’s visual-analysis benchmark, trailing only Anthropic’s Claude Fable 5, according to The Next Web.
Alibaba also said the model completed a software-engineering project during a 16-day autonomous run without human supervision — a claim The Next Web notes is Alibaba’s own and has not been independently tested. On the capabilities side, Alibaba said the multimodal model can process long documents, television series, and live streams to build searchable knowledge bases, and that it can recreate software applications from screenshots, generate interactive games and educational animations, and convert two-dimensional floor plans into 3D visualizations, according to SCMP.
The announcement moved Alibaba’s stock: the company’s Hong Kong-listed shares rose 7 percent following the release, closing at HK$125.20 the same day, SCMP reports.
Qwen3.8-Max follows Qwen3.7-Max, which The Machine Herald reported Alibaba unveiled at its Cloud Summit as a long-horizon agent model. SCMP notes the new release also signals Alibaba’s entry into a recent round of powerful releases by Chinese developers aggressively narrowing the gap with leading US labs.
What We Don’t Know
Alibaba has not published an exact date for the open-weights release beyond the general timeframe reported by outlets covering the announcement, nor has it disclosed full training details for Qwen3.8-Max. The company’s claim of a 16-day autonomous software-engineering run has not been independently verified, and it remains unclear how Qwen3.8-Max’s real-world performance will hold up once developers gain broader hands-on access to the open-weight release.
Analysis
The release continues a rapid-fire cadence among Chinese AI labs this year, with Alibaba, Moonshot AI, and others trading claims to the top of leaderboards in quick succession. Alibaba’s decision to open-source a Max-class flagship again, after a stretch of closed-weight releases, suggests the company is again prioritizing developer mindshare and Alibaba Cloud workload capture over keeping its most capable model proprietary — a strategy Chinese developers have increasingly leaned on as a point of differentiation from US labs that keep architecture and parameter counts undisclosed.