Alibaba Releases QwenLong-L1.5 Enhancing Long-Context Reasoning

Alibaba Tongyi Lab released QwenLong-L1.5, a long-context reasoning LLM post-trained from Qwen3-30B-A3B-Thinking that supports reasoning beyond its 256K token native window. It combines long-context data synthesis, Adaptive Entropy-Controlled Policy Optimization (AEPO), and a multi-stage memory fusion framework, and the article includes instructions to run inference on DigitalOcean GPU Droplets with recommended H100/A100 GPUs and 80GB VRAM.