Greg Brockman 转发 @arcprize 的评测称 OpenAI 的 GPT-6 Astra 在 ARC-AGI-3 上取得 SOTA，他称该基准已饱和。
Astra 标准 harness 得分 63%，经新的 Provider Adapter harness 达 99%，在 96% 的 ARC-AGI-3 关卡上超越人类表现；
排行榜图还显示更高推理层级通常成本更低，因为 Astra 用更少动作通关，减少模型调用和 token 数。
AI News / Industry
Back to AI news
Greg Brockman retweeted: GPT-6 Astra reached SOTA on ARC-AGI-3, benchmarking is approaching saturation
Greg Brockman retweeted @arcprize's evaluation stating that OpenAI's GPT-6 Astra achieved SOTA on ARC-AGI-3, and he mentioned that this benchmark is alread...
Today AI Intelligence Brief
Greg Brockman retweeted @arcprize's evaluation stating that OpenAI's GPT-6 Astra achieved SOTA on
ARC-AGI-3, and he mentioned that this benchmark is already saturated. Astra scored 63% on the standard harness, and with the new Provider Adapter harness, it reached 99%, surpassing human performance on 96% of ARC-AGI-3 levels. The leaderboard chart also shows that higher reasoning levels usually cost less, because Astra completes levels with fewer actions, reducing model calls and token consumption.
动态正文 · AI 整理
Intelligence Assessment
Aioga 编辑摘要
据 Greg Brockman 转发的 @arcprize 评测，材料称 OpenAI 的 GPT-6 Astra 在 ARC-AGI-3 取得 SOTA；标准 harness 得分 63%，Provider Adapter harness 得分 99%，并在 96% 关卡上超过人类表现。
背景分析
材料同时称，排行榜图显示更高推理层级通常成本更低，原因是 Astra 以更少动作通关，从而减少模型调用和 token 数；Brockman 还称该基准已饱和。
Aioga 观点
Aioga 判断：这则信息的核心不只是榜单分数差异，也包括评测 harness 变化带来的结果差异。由于材料仅提供转述与摘要，SOTA、饱和及成本关系仍应结合完整评测规则核验。
影响与后续
可能影响：若相关评测结果可复现，ARC-AGI-3 的比较将需要同时关注 harness、动作数量、模型调用和 token 消耗；但单条材料不足以证明通用能力提升，也不代表所有任务成本都会下降。 后续观察：建议核对 ARC-AGI-3 排行榜、标准与 Provider Adapter harness 的定义、样本覆盖和复测结果，并持续观察“基准饱和”判断是否得到后续公开数据支持。
Source and Copyright
The readable text on this page was extracted from the public source and organized with attribution, publication time and the original link. Copyright remains with the original author and publisher.
Ingestion channel: Summary aggregation · Source domain: x.com
Source: @gdb)
Original link: Open original source
Aioga archive: Open intelligence page
Content record: social-summary · Updated: 2026-09-03T21:46:00.000Z
API RELAY · DEVELOPER INFRASTRUCTURE
API 中转站
统一接入主流 AI 模型 API，为开发、测试与生产环境提供稳定调用入口。
立即访问 api.w173.com
Scan to open this article