Sapient Intelligence 发布自主研发系统 PRAXIST Beta 并宣布即日起开源，在完整 75 项任务的 MLE-bench 上，使用 deepseek-v4-pro 的 PRAXIST 取得 49 枚金牌、总计 60 枚奖牌，Claude Opus 4.8 为 34 枚金牌、55 枚奖牌。
MLE-bench 的对比数据展示了自主研发系统的成绩，共享研究图谱的复用机制是其中可关注的设计。
Sapient Intelligence released PRAXIST Beta, an autonomous R&D system that carries experimental evidence across generations.
On the full 75-task MLE-bench:
- Sapient's PRAXIST got 49 gold medals and 60 medals overall for PRAXIST using deepseek-v4-pro.
- Claude Opus 4.8 reached 34 gold and 55 total
And PRAXIST is open source starting today
Instead of keeping only the best candidate, PRAXIST runs parallel Research Peers and stores findings, failures, evidence strength, and lineage in a shared research graph.
That graph lets later experiments reuse validated mechanisms, retest weak claims, and combine useful discoveries from different research branches.
来源：@rohanpaul_ai · x.com