OpenAI LifeVal 图表：GPT-6 Astra 得分 80.6%，较 GPT-5.2 提升 31 个百分点
OpenAI 自己发的 LifeVal 榜单，GPT-6 Astra 比 GPT-5.2 涨了 31 个点，不过任务集没公开，看看就好
OpenAI 的 LifeVal 基准显示，GPT-6 Astra 在日常生活任务上的平均奖励达 80.6%，而 GPT-5.2 为 49.3%。示例任务中，搬家后幼儿园筛选得分 93.1%，洪灾志愿者计划得分 98%，一项符合预算和通勤要求的 Luzerne County 找房任务拿到 100%。任务集来自时间使用调查以及对 3995 人的问卷，覆盖育儿、健康、教育、财务、购物和政务服务。需要注意的是，这是 OpenAI 自家的评测，只测 OpenAI 模型，完整任务集未公开。
OpenAI's LifeVal chart puts GPT-6 Astra at 80.6% average reward on everyday tasks, up from 49.3% on GPT-5.2. The example runs it shows are higher still: 93.1% on kindergarten shortlists after a move, 98% on flood volunteer plans, and 100% on a Luzerne County home search that fits budget and commute.
It's OpenAI's own eval, on OpenAI models only, and the full task set isn't public. Tasks came from time-use surveys plus a poll of 3,995 people across caregiving, health, education, money, shopping and government services.
