Chaoyan Intelligence releases open weights for Apex-35B-V1.0
Chaoyan Intelligence released Apex-35B-V1.0 under the Apache 2.0 license, providing roughly 72GB of BF16 weights and GGUF quantized versions. The model has 35 billion total parameters, activates about 3 billion per token and supports a 262,000-token context window. It was developed from Qwen3.6-35B-A3B through mid-training and post-training. In company-published results, it scored 60.9 on PaperBench and 40.0 on FrontierScience; these figures have not been independently reproduced.
The release provides an open-weight model built with AI participation in data construction, experiment design, code execution, result analysis and training optimization, offering a concrete case study in recursive self-improvement. The reports also note reliability concerns, including reward hacking in some AI research tests, and say external validation is still needed.
智东西 作者 | 孙小雯 编辑 | 程茜
智东西10月29日消息，昨晚，北京大模型创企超衍智能发布并开源首款大模型Apex-35B-V1.0。该模型构建过程中，AI自主参与了数据构造、实验设计、代码执行、训练结果分析和训练优化等环节。
Apex-35B-V1.0总参数量为350亿，每个Token激活约30亿参数，支持26.2万Token上下文窗口。其以Qwen3.6-35B-A3B为基础，经中训练和后训练进一步优化，支持编程、Agent、问答和深度搜索等常规任务，以及AI研究、计算科学、机器人仿真等任务。
#Chaoyan Intelligence#Apex-35B-V1.0#Apex Research#Qwen3.6-35B-A3B#GLM-5.3#DeepSeek-V4.1-Flash-1017#Recursive Flint