Alibaba has released Qwen3.8-2.4T-A95B, marking the first open-weight release of a model in its Qwen-Max class. This new model boasts 2.4 trillion total parameters, with 95 billion active parameters per forward pass, utilizing a hybrid architecture of Gated DeltaNet and standard attention. Qwen3.8-Max demonstrates strong performance across various benchmarks, particularly excelling in PaperBench for academic understanding and coding tasks, while also offering competitive pricing at $2/$6 per million tokens for input/output. AI
IMPACT This release provides a powerful, open-weight model with a unique architecture, potentially driving innovation in agentic tasks and long-context applications.
RANK_REASON Frontier-lab model release with system card and benchmarks. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
- Alibaba
- Claude Fable 5
- Gated DeltaNet
- GPT-5.6 Sol
- Mixture of Experts
- NVIDIA
- Opus 4.8
- PaperBench
- Qwen3.8-2.4T-A95B
- QwenCloud API
- Qwen-Max
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
