Alibaba launched Qwen4-Max on November 16, 2026, as the flagship model in its Qwen4 series, built around a sparse Mixture-of-Experts architecture with a 2.4-trillion-parameter backbone. The multimodal model handles text, images and video and supports context windows of up to 1 million tokens. On the DeepSWE v1.1 coding benchmark, Alibaba said Qwen4-Max scored 75.7, ahead of Claude Opus 5.5 and competitive with Kimi K3.1 Pro, although independent rankings place it behind some rivals while recognizing its lower cost per task. Alibaba said the model uses approximately 50% fewer tokens than Qwen3.8-Max and reduces average task costs by more than 60%. It is priced at $1.50 per million input tokens and $4.50 per million output tokens, with research findings putting its cost per task at roughly $3.27. Developers can access it through the QwenCloud API, using protocols compatible with OpenAI and Anthropic systems. A Token Plan preview runs from November 11 through November 25, while smaller Qwen4 models are planned as open-weight releases and Qwen4-Max will remain closed. The launch emphasizes efficiency and practical coding, healthcare, finance and legal applications, while third-party testing remains important because benchmark results vary by task.