Skip to content
AIBRAINWORX / THE INDEPENDENT AI LABRESEARCH · HARDWARE · OPEN INTELLIGENCE
AI News / Model Watch

Alibaba launches Qwen4-Max as a 1M-token closed API flagship

The November 16 release pairs a 2.4T-parameter Mixture-of-Experts backbone with 71B active parameters per token, while smaller open-weight Qwen 4 variants remain forthcoming.

Official Alibaba Qwen profile mark
Alibaba Qwen, published Hugging Face profile icon; identification only, no endorsement.
Event / context: Alibaba launched Qwen4-Max generally on November 16, 2026, after a limited preview starting November 11.. Source check: October 28, 2026.

Alibaba has launched Qwen4-Max, its next-generation flagship model, moving the system from a November 11 limited preview to general API availability on November 16, 2026. Unlike several earlier Qwen cycles that shipped open-weight checkpoints alongside hosted endpoints, Qwen4-Max debuts strictly as a closed cloud API.

Architecture, context window, and pricing

Published technical specifications describe a sparse Mixture-of-Experts (MoE) architecture with a 2.4-trillion-parameter backbone that activates roughly 71 billion parameters per token, paired with 328 billion N-gram embedding parameters. The model supports a 1-million-token native context window across text, image, and video inputs, with built-in reasoning enabled by default and adjustable reasoning-effort controls.

On QwenCloud international pricing, list rates are $1.50 per million input tokens, $4.50 per million output tokens, and $0.075 per million implicit cache-hit input tokens—roughly 25 percent below Qwen3.8-Max list pricing. OpenAI- and Anthropic-compatible API endpoints are supported alongside Alibaba's QwenWork, Qoder, and Qwen Studio applications.

What remains open and what is missing

Alibaba indicated that smaller open-weight Qwen 4 variants will follow, but no weights, local-serving checkpoints, or open-source license terms exist for Qwen4-Max itself. Developers looking for local deployment remain on prior open-weight releases such as Qwen 3.8 27B under Apache 2.0. AiBrainWorX has not independently benchmarked Qwen4-Max; performance and latency claims rest on publisher documentation.

Original reporting; primary and corroborating sources reviewed November 16, 2026. No independent hands-on test claimed.

Follow your curiosity.

Pass the signal.