1
00:00:00,000 --> 00:00:03,480
The progress over the recent scale is quite amazing.

2
00:00:03,680 --> 00:00:09,080
Our very first reasoning model was like a top one millionth competitive programmer in the world.

3
00:00:09,560 --> 00:00:10,720
People thought that was very impressive.

4
00:00:10,880 --> 00:00:14,100
It's like, wow, an AI, it's the millionth best people that do this.

5
00:00:14,120 --> 00:00:14,660
That's pretty good.

6
00:00:15,700 --> 00:00:19,640
We then had a model that got to like a top 10,000.

7
00:00:21,240 --> 00:00:27,820
03, which we talked about publicly in December, is the 175th best competitive programmer in the world.

8
00:00:27,820 --> 00:00:33,220
I think our internal benchmark is now around 50, and maybe we'll hit number one by the end of this year.

9
00:00:33,840 --> 00:00:40,760
So that's like an amazing rate of scale for more compute in this new paradigm, and we don't see any signs of that stopping.