上下文窗口1M
最大输出32K
知识截止—
发布日期2026-04-08
系列muse
最后更新2026-07-09
输入模态文本图像PDF视频
输出模态文本
能力
推理工具调用结构化输出温度调节文件附件
基准跑分
SWE-Bench Pro 61.5 (resolve rate) · Terminal-Bench 80 (success rate) · DeepSWE 53.3 (resolve rate) · MCP Atlas 88.1 (success rate) · JobBench 54.7 (success rate) · Toolathlon-Verified 75.6 (success rate) · Humanity's Last Exam 62.1 (accuracy) · OSWorld-Verified 80.8 (success rate) · Finance Agent 57.2 (accuracy) · CharXiv Reasoning 88.4 (accuracy) · BabyVision 76.3 (accuracy)