[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.]

![chatgpt21 Avatar](https://lunarcrush.com/gi/w:24/cr:twitter::1613004034096381952.png) Chris [@chatgpt21](/creator/twitter/chatgpt21) on x 12.2K followers
Created: 2025-07-17 04:10:40 UTC

This is a huge win.

First autonomous coding victory
An OpenAI agent entered a 10-hour AtCoder Heuristic Finals and beat yosupo (Codeforces #76) with zero human tweaks. That’s the first time an LLM-based system has outperformed a top-tier human in a live, long-horizon programming contest.

This might be True recursive self-improvement from Open AI

The agent didn’t just follow prompts it:

> Started with a simple heuristic

> Monitored its own scores and failures

> Tweaked its code to learn better wall placement strategies

> Reallocated resources more effectively

> Optimized path-finding routines
…all without new human guidance. That feedback→rewrite loop is classic “recursive self-improvement”

![](https://pbs.twimg.com/media/GwCBUcJXYAAeWOf.jpg)

XXXXXX engagements

![Engagements Line Chart](https://lunarcrush.com/gi/w:600/p:tweet::1945697637669495184/c:line.svg)

**Related Topics**
[coins ai](/topic/coins-ai)
[finals](/topic/finals)
[open ai](/topic/open-ai)

[Post Link](https://x.com/chatgpt21/status/1945697637669495184)