Alibaba releases Qwen3.8-Max AI model with 2.4 trillion parameters, open weights to follow next week

Alibaba Group Holding has launched Qwen3.8-Max, a flagship AI model boasting 2.4 trillion parameters. Operating on a sparse Mixture-of-Experts architecture, it activates 95 billion parameters per step to optimize efficiency. Alibaba claims the model rivals top US systems like Anthropic's Claude Fable 5, particularly in visual tasks. Priced at $2 per million input and $6 per million output tokens, the model is currently in preview, with open weights scheduled for public release in the week of August 10, 2026.

Qwen 3.8 Max announcement

• ▪Alibaba Group Holding plans to release the open weights for Qwen3.8-Max and a smaller Qwen3.8-27B model during the week starting August 10, 2026
• ▪Alibaba Group Holding released its flagship artificial intelligence model, Qwen3.8-Max, on August 3, 2026, which is available via the QwenCloud platform

Sparse MoE architecture specifications

• ▪Qwen3.8-Max activates approximately 95 billion parameters per processing step to reduce computational costs and latency
• ▪Qwen3.8-Max is built on a sparse Mixture-of-Experts architecture containing 2.4 trillion total parameters
• ▪Qwen3.8-Max supports a context window of up to one million tokens and can process text, images, and video

Preview endpoint availability status

• ▪Alibaba Group Holding previewed a pre-release version of Qwen3.8-Max on July 19, 2026, at the World AI Conference in Shanghai
• ▪The Qwen3.8-Max preview endpoint is accessible to developers through Alibaba Cloud's Model Studio APIs, QwenWork, Token Plan, and Qoder

Unverified benchmark claims

• ▪Alibaba Group Holding claims Qwen3.8-Max ranks second globally behind Anthropic's Claude Fable 5 on Arena.AI's leaderboard for visual understanding
• ▪In an independent architecture evaluation, Qwen3.8-Max preview scored 80 out of 100, while Moonshot AI's Kimi K3 scored 83

Token Plan subscription pricing

• ▪Alibaba Group Holding charges $0.25 per million tokens for implicit input caching of Qwen3.8-Max
• ▪Alibaba Group Holding has priced Qwen3.8-Max API access at $2 per million input tokens and $6 per million output tokens

Qwen generation comparison

• ▪Qwen3.8-Max is priced higher than its predecessor, Qwen3.7-Max, which costs $1.25 per million input tokens and $3.75 per million output tokens
• ▪Unlike Qwen3.7-Max, which focused on agentic workloads, Qwen3.8-Max introduces native multimodal input capabilities