This repository contains model weights and configuration files for the Qwen3.8 post-trained model in Hugging Face Transformers format. The official Qwen API service is provided by Qwen Cloud, with Qwen3.8-Max offering vision input, non-thinking support, and 1M context length.

Following the Qwen3.5 and Qwen3.6 series, Qwen3.8 is the most capable generation in the Qwen open-model family. It brings a Qwen-Max-class model to open release, delivering substantial gains across coding, professional work, research, and long-horizon agentic tasks. Key features include stronger autonomous planning, broader downstream compatibility, and flexible thinking control via reasoning_effort.

The model has 2.4T total parameters with 95B activated, context length of 262,144 natively extensible to 1,010,000 tokens, and uses a mixture of experts architecture with 512 experts.

Benchmark results show Qwen3.8-Max outperforming previous models on coding agents (Terminal Bench 2.1), general agent tasks, and general capabilities like GPQA Diamond. For streamlined integration, Qwen recommends using the API or frameworks like SGLang, vLLM, or Token Speed for production workloads.