Alibaba's Qwen 3.8 Max Puts 2.4 Trillion Parameters on a Public API and Promises Open Weights by August 10
Alibaba launched Qwen 3.8 Max on August 3, 2026. As the flagship release of the new Qwen 3.8 series, it is their largest model to date: 2.4 trillion total parameters, 95 billion active per token, using a sparse Mixture-of-Experts architecture. The release brings the highly anticipated qwen open weights roadmap to the forefront, offering a model that accepts text, image, video, and document inputs in a single request and supports a 1-million-token context window. It is live now via Alibaba Cloud Model Studio and the QwenWork platform.
API pricing is set at 12 RMB per million input tokens and 36 RMB per million output tokens — roughly comparable to mid-tier US frontier model pricing on a purchasing-power-adjusted basis. The bigger near-term story is on the open-weights side: Alibaba announced that Qwen 3.8 Max will be the first "Max-class" model in the Qwen family to have weights released publicly, with that release scheduled for the week of August 10. That would make it one of the largest open-weight models available (and the largest open weight LLM to date), exceeding Kimi K3's 1.4-trillion-parameter weight release from late July.
Alibaba's internal evaluations position the model as second only to Claude Fable 5 on benchmark composites, outperforming [GPT-5.6 Sol](/models/gpt-5-6-sol) and Gemini 3.1 Pro on coding and multimodal understanding tasks, while trailing in some general-purpose reasoning categories — a pattern consistent with how flagship chinese open source llm variants have typically landed relative to western frontier labs. Independently, Qwen 3.8 Max has already appeared on the LMSYS Chatbot Arena leaderboard, where it ranks first among Chinese models on text tasks and second globally in the multimodal/visual analysis category — the only third-party signal available at launch.
The release lands a week after Kimi K3's open weights went public, and two days after the EU AI Act's GPAI transparency rules took effect — a regulatory backdrop that may shape how Alibaba positions open-weight access in European markets. This release highlights Alibaba's open source AI roadmap, cementing their position in the open weights ecosystem. For context on where the broader Qwen family sits, see our earlier coverage of Qwen 3.7 Flash, the cost-tier counterpart that launched at $0.03 per million tokens in July. Qwen 3.8 Max is now clearly the flagship generation; the 3.7 series is the volume tier.
