Particle.news

Alibaba Launches Qwen3.8-Max, a 2.4 Trillion-Parameter Multimodal Model

Paid previews are live with open weights due within days, signaling lower developer costs and a rush of independent testing.

Overview

  • Alibaba unveiled Qwen3.8-Max on Monday, Aug. 3, 2026, presenting it as its largest model and a multimodal system that handles text, images and long documents.
  • The company says the model has 2.4 trillion total parameters, but it uses a mixture-of-experts design that Alibaba reports activates about 95 billion parameters per request to cut compute and latency.
  • Early paid access is available through platforms such as Token Plan and Qoder with promotional discounts up to 90 percent, and Alibaba committed to releasing the model's open weights within days.
  • Performance claims so far rely on Alibaba’s internal benchmarks and leaderboard snapshots, leaving third-party validation and a detailed active-parameter breakdown as the next tests of its true capability.
  • The launch pushed Alibaba shares higher and could pressure proprietary AI pricing while boosting interest in lower-cost inference options and adding to China’s push for AI self-reliance despite export-control limits.