X：Yuchen Jin (@Yuchenj_UW)· @Yuchenj_UW·· 6 天前AI 评分
Databricks 用 GPT-6 Astra 与 Opus 5 智能体登顶 NVIDIA SOL-ExecBench 内核榜四赛道
We @databricks now rank #1 on NVIDIA’s SOL-ExecBench kernel leaderboard, across all 4 tracks. We put GPT-6 Astra and Opus 5 in a self-hillclimbing loop, and the agents beat the top GPU kernel engineers. Total toten spend: ~$70K, far less than what kernel engineers get paid. One learning: GPT-6 Astra and Opus 5 are still much better than OSS models at writing GPU kernels.
Databricks 在 NVIDIA 的 SOL-ExecBench 内核排行榜上排名第一，覆盖全部 4 个赛道。团队把 GPT-6 Astra 和 Opus 5 放进自我爬坡循环，智能体表现超过顶尖 GPU 内核工程师，总 token 花费约 7 万美元，远低于内核工程师的薪酬。作者提到一个发现，GPT-6 Astra 和 Opus 5 在编写 GPU 内核上仍明显优于开源模型。
来源：X：Yuchen Jin (@Yuchenj_UW) · x.lingyaoai.com