OpenAI's experimental AI model wins IMO gold, teases upcoming GPT-5 release

卮言
1

OpenAI宣布，其最新实验性推理模型在人工智能领域取得重大突破，首次达到国际数学奥林匹克竞赛(IMO)金牌水平。该模型在2025年IMO题目评测中，严格按照人类参赛规则进行：两场各4.5小时的考试，不使用任何工具或网络，仅凭阅读官方题目并撰写自然语言证明。最终成功解决6道题目中的5道，获得35分(满分42分)，达到金牌标准。

IMO题目要求持续创造性思维，从简单数学题的0.1分钟推理提升至IMO的100分钟复杂推理，代表了AI在数学证明能力上的质的飞跃。

OpenAI表示即将发布GPT-5，但这项IMO级别的数学能力暂不会在数月内向公众开放。值得一提的是，在本届IMO比赛中，中国队以231分的总成绩夺得团体冠军，六名选手全部获得金牌。

Last edited by @suen 2025-07-19T13:12:31Z

2 Likes
2

https://x.com/alexwei_/status/1946477742855532918

1 Like
3

https://x.com/alexwei_/status/1946477745627934979

1 Like
4

https://x.com/alexwei_/status/1946477762023412135

1 Like
5

https://x.com/zhansheng/status/1946492965796581827

1 Like
6

https://x.com/SebastienBubeck/status/1946577650405056722

1 Like
7

https://x.com/sama/status/1946569252296929727?s=46

1 Like
9
1 Like
10

https://x.com/zjasper666/status/1946335438228361372?s=46

1 Like
11

https://x.com/ai_for_success/status/1946984782178709719

1 Like
12

https://x.com/lmthang/status/1946960256439058844

1 Like
13

https://x.com/lmthang/status/1947334216544621041?s=46

1 Like
14
1 Like
15

Google’s is recognized, OAI is not. And given OAI’s tradition, it’s hard for us to see that model. Do you like my O3 trained on the problem?

16

https://x.com/askalphaxiv/status/1948095637565964354?s=46

1 Like