Qwen3.8-Max, released August 3, 2026, is described by as the largest and most capable model in the Qwen family and the first time Alibaba has open-sourced a Max-level flagship model, with weights scheduled for Hugging Face and ModelScope. A separate model, Qwen3-Max-Preview, carries more than 1 trillion parameters and is accessible through Alibaba's paid API, Qwen Chat, OpenRouter, and AnyCoder, with no open-source license. A 2.4T-parameter open-weight mixture-of-experts model, Qwen3.8-2.4T-A95B, was released on Hugging Face. The Qwen3 family includes two MoE models (Qwen3-235B-A22B and Qwen3-30B-A3B) and six dense models from 0.6B to 32B parameters, all under the Apache 2.0 license, with benchmark scores of 94.39 on GSM8K, 88.87 on BBH, and 87.81 on MMLU. On the Arena leaderboard (formerly LMArena, rebranded January 2026), Claude Fable 5 leads with approximately 1525 ELO. One tracker shows a Qwen3 235B thinking model at rank 52 with Elo 1419 and Claude Opus 4.1 at rank 56 with Elo 1418. A separate source describes the Qwen3 235B-Instruct as ranked 8th on LMArena, tied with Claude Opus 4, with a 1T variant outperforming Kimi K2, DeepSeek 3.1, and Claude Opus 4 on all major benchmarks. Qwen3-8B-Max climbed from rank 22 to rank 4 on the Legal Research Bench, nearly doubling its score in under three months.