Overview
- Alibaba formally revealed Qwen3.8‑Max, which it unveiled on Monday, Aug. 3, as a 2.4 trillion-parameter, multimodal model that uses a mixture-of-experts design and supports very long context windows.
- The company plans to publish the model's weights for developers to download, enabling self-hosting and local fine-tuning while activating roughly 95 billion parameters during inference.
- Independent agentic benchmarks and indexes have placed Qwen3.8‑Max at or near the top on multi-step reasoning, coding, and tool-use tasks, showing strong performance on long-horizon and multimodal workloads.
- Reuters and other outlets report Alibaba will require major commercial users to enter revenue-sharing or commercial agreements for large-scale deployments, though exact thresholds and rates remain unclear.
- Alibaba is pairing aggressive per-token pricing and enterprise integrations with this open-weight release to funnel usage toward Alibaba Cloud, which could lower costs for developers but shift where major customers pay for compute and hosting.