上下文窗口1.1M
最大输出128K
知识截止2026-02-16
发布日期2026-07-09
系列gpt-terra
最后更新2026-07-09
输入模态文本图像PDF
输出模态文本
能力
推理工具调用结构化输出文件附件
基准跑分
SWE-Bench Pro 63.4 (resolve rate) · Terminal-Bench 87.4 (success rate) · DeepSWE 69.6 (resolve rate) · GPQA Diamond 92.9 (accuracy) · FrontierMath 84.9 (accuracy) · BrowseComp 87.5 (accuracy) · OSWorld 50.2 (success rate) · MMMU Pro 80.7 (accuracy) · Agents' Last Exam 50.4 · Toolathlon 53.1 (success rate) · Artificial Analysis Intelligence Index 55 (index score) · Artificial Analysis Coding Agent Index 77.4 (index score)