Alibaba’s Qwen3.8-Max, a 2.4-trillion-parameter sparse MoE large language model activating roughly 95 billion parameters with a 1-million-token context window, launched in early August 2026 and remains the current flagship. Trader sentiment for a 3.9+ successor centers on Alibaba’s rapid release cadence—nearly one model every seven days historically—combined with the late-October Apsara Conference announcement that Qwen 4 is already in training and a roadmap exists for 4.5 and 5 series scaling to 5–10 trillion parameters. Recent recursive self-improvement gains on Qwen3.8-Max and multimodal updates have sustained momentum, while competitive pressure from other Chinese labs keeps timelines tight. Markets currently imply an 82% chance of a 3.9+ release by November 30.