Qwen3.8-Max-Preview
Alibaba·
LLMsflagshipcloud
Overview
Preview of Alibaba's flagship Max model, unveiled at WAIC Shanghai. 2.4T-parameter sparse Mixture-of-Experts, natively multimodal (text, images, video, documents) with a 1M-token context window (inherited from Qwen3.7-Max). Active-parameter count, full benchmark suite and license were not disclosed at preview. Runs at 10% of standard pricing during the preview via Token Plan / Qoder; an open-weight release was announced as forthcoming.
Capabilities and innovations
2.4T total parameters (sparse MoE)1M context windowNatively multimodal (text, image, video, docs)Preview via Token Plan / QoderOpen weights planned2.4T Sparse MoEMultimodal Text/Image/Video/DocsOpen-Weight Max Release Planned
Benchmarks
No benchmark values are recorded for this model.
- GPQA Diamond: independent evaluation pending — Preview at WAIC 2026-07-19; benchmark table not yet published, no independent evals
- Humanity’s Last Exam (no tools): independent evaluation pending
- SWE-bench Pro: independent evaluation pending
- MMLU-Pro: independent evaluation pending
- Terminal-Bench 2.x: independent evaluation pending
- MMMU-Pro: independent evaluation pending
Architecture and hardware
- Parameters
- 2400B total (MoE)
Links
More from Alibaba
Data curated by AI Model Timeline. See the methodology for admission criteria, benchmark eras and source priority.