What is the parameter count of Qwen3.8?

Qwen3.8 has 2.4 trillion parameters, making it the second Chinese model after Kimi K3 to exceed 2 trillion.

What is the pricing for Qwen AI Platform?

In China, input is RMB 12 and output RMB 36 per million tokens. Internationally, it's 40% and 24% of Opus5.

What are the features of Qwen Office?

The first product to integrate desktop, cloud, and collaboration agents, connected with DingTalk for efficiency.

How does Qwen3.8 perform in benchmarks?

Ranked second only to Anthropic's Claude in Arena, with 1.5x efficiency gain in agentic reasoning.

Is Qwen3.8 open-sourced?

Qwen3.8-Max and Qwen3.8-27B will be open-sourced next week.

Following Kimi K3, Another 2-Trillion-Parameter Breakthrough! Alibaba Launches New-Gen AI Model Qwen3.8-MAX, Stock Surges Over 7%

Published Aug 3, 2026 2:50 PM ・ Updated Aug 3, 2026 3:08 PM ・ 5 min read ・ Source: PR Times

Alibaba unveiled its next-generation large model Qwen3.8 with 2.4 trillion parameters, boosting coding and office productivity. The stock rose over 7%, and Qwen3.8 ranks second only to Anthropic's Claude in global benchmarks.

Alibaba's Hong Kong-listed shares surged over 7% in morning trading today (3rd), continuing gains in the afternoon and briefly reaching HK$126 per share. This rally was primarily driven by Alibaba's official launch of its new-generation foundational large model, Qwen3.8, which boasts a total of 2.4 trillion parameters and significantly enhanced capabilities in coding and professional office tasks (Cowork). Following Kimi K3, Qwen3.8 becomes the latest domestically developed large model in China to surpass the two-trillion-parameter threshold.

In today's authoritative third-party leaderboard Arena, Qwen3.8 ranks just behind Anthropic's Claude series, placing it among the top-tier global large models in overall performance.

Currently, the Qwen3.8 API is live on the Qwen AI platform and integrated with Alibaba's newly launched agent product, 'Qwen Office.' Qwen3.8-Max is scheduled to be open-sourced next week, alongside the open-sourcing of Qwen3.8-27B.

Notably, Alibaba's Zhenwu M890 super-node has successfully adapted to Qwen3.8. Through full-stack optimization across chips, cloud platforms, and the Qwen large model, inference efficiency has been significantly improved and inference costs reduced. In agentic reasoning scenarios, performance gains of up to 1.5x have been achieved.

Importantly, Alibaba also announced the public beta launch of its enterprise-grade agent product 'Qwen Office' today. Users can now experience Qwen3.8 through this product. The new 'Qwen Office' integrates three products—QoderWork, MuleRun, and Wukong—making it the industry's first solution to support desktop agents, cloud agents, and enterprise collaboration agents simultaneously. It has already begun integrating with DingTalk IM and will fully connect to real enterprise databases and workflows in the future, helping businesses boost efficiency.

Starting today, global developers can access Qwen3.8 API services via the Qwen AI platform. In China, the pricing is RMB 12 per million input tokens and RMB 36 for output, with implicit cache hits costing only RMB 1.5. Internationally, input and output prices are just 40% and 24% of Opus5's rates, respectively, offering highly competitive pricing.

FACT BOX

Source: PR Times
Category: New Product
Organizations: Anthropic
Products / services: Qwen3.8-MAX / QoderWork

Following Kimi K3, Another 2-Trillion-Parameter Breakthrough! Alibaba Launches New-Gen AI Model Qwen3.8-MAX, Stock Surges Over 7%

⚡ Key Points

FACT BOX

Editorial & Verification Standards

FAQ

Cite this article — HOW TO CITE

AI CRAWLER ACTIVITY