Qwen3.8 Max (NovitaAI)

Qwen3.8 Max is described by Alibaba as a 2.4-trillion-parameter mixture-of-experts flagship built upon the architectural foundation of Qwen 3.5, with the official announcement positioning it as the most capable model in the Qwen family to date. The release notes emphasize comprehensive improvements in coding and general work tasks, and notably mark the first time a Qwen-Max-class model will have its weights open-sourced, with the open release slated to follow the initial closed launch. This lineage and scale story positions the model as a step forward in the Qwen family's progression toward increasingly large MoE architectures optimized for complex reasoning and software-style tasks.

Through the Novita AI gateway, Qwen3.8 Max is offered as a chat model with a one-the cataloged API limit and a 131,072-token maximum output, served via an OpenAI-compatible Chat Completions endpoint that supports function/tool calling, structured outputs, JSON mode, streaming, fine-tuning, and batch processing. That combination of a very long context, generous output budget, and tool/structured-output integration makes it a practical fit for long-document analysis, agentic coding workflows, and complex multi-step knowledge work where the model needs to ingest large codebases or reports and emit richly formatted responses. The MoE backbone keeps activation costs in check relative to total parameter count, while the announced open-weights follow-up signals that customization and on-premise deployment will become viable shortly after launch.