Qwen 3.8 Max Lands at 2.4T Parameters / 95B Active With 87.3% SWE-bench — but the Open Weights Still Aren't on Hugging Face and the License Reportedly Excludes the US, EU, UK and Korea
Latent Space (AINews)·high signal
Alibaba announced Qwen 3.8 Max on August 3, 2026: a 2.4T-parameter sparse MoE with ~95B active per token, 1M context, 128k max output, priced at $2/M input and $6/M output with $0.25/M cached. It scores 87.3% on SWE-bench, 67.4 on Terminal-Bench 2.1 (up from 61.0 for Qwen 3.7 Max), #4 on Frontend Code Arena at 1,668 Elo and #2 on Vals Index among open-weight models at 66.1. The catch AINews flagged: weights for both Max and the companion Qwen3.8-27B were promised "next week" on Hugging Face/ModelScope with no license named at announcement, and community reports say usage is restricted in the US, EU, UK and Korea — a sharp break from the Apache 2.0 pattern of prior Qwen lines.