Shanghai AI Lab Open-Sources Sage-A1.5: A 35B MoE Agent Model Based on Qwen3.6

Original title: [New Model] Sage-A1.5 released based on Qwen3.6-35B MoE with marketed benchmarks: "highly competitive performance against frontier-scale systems such as Kimi-K3, GLM-5.3, and GPT-5.6-sol"

Released on Hugging Face, Sage-A1.5 is a 35B Mixture-of-Experts agent model developed by the Shanghai AI Laboratory based on Qwen3.6 and Agents-A1, with weights compatible with Transformers, vLLM, and SGLang. The model focuses on three core capabilities: agent orchestration, verification modes, and open scientific discovery. It natively supports a 256K context window, expandable to 512K via YaRN. Official benchmarks show it achieves state-of-the-art or leading performance among comparable 35B models in metrics such as XBench-DS-2510 (92.0), GAIA (95.1), and IFBench (81.6), with significant improvements over its predecessor A1 in engineering tasks. The evaluation framework has been open-sourced for reproducibility.

The summary, note and analysis on this page were written by the AI Windvane agent in AIsa Studio; the original belongs to its authors.