We will hold AtCoder World Tour Finals 2026 Algorithm.
- Contest URL: https://atcoder.jp/contests/awtf2026algo
- Mirror Contest URL: https://atcoder.jp/contests/awtf2026algo-open
- Start Time: https://www.timeanddate.com/worldclock/fixedtime.html?iso=20260709T1100&p1=248
- Duration: 420 minutes
- Number of Tasks: 5
- Writer: maroonrk
- Tester: hos.lyric, tatyam, maspy, yosupo, sigma425
- Rated range: Unrated
The point values will be 900-900-1500-2500-2500. We don't freeze the standings in the contest.
AtCoder World Tour Finals 2026 is an international onsite contest. We have invited top-runners of AtCoder Race Ranking 2025 to Tokyo. This year, as an exhibition match, an AI agent from OpenAI will also participate in the contest. On the contest day, we will stream the contest on our YouTube channel with Petr. For more details, please visit this page ( https://info.atcoder.jp/more/contents/awtf/2026 ).
The mirror contest is Unrated and the standings is hidden. Other rules are the same as AGC.
P.S. This is not about Algorithm contest, but we also have a Heuristic Contest in the Finals and Psyho will do the commentary stream tomorrow, so please check it out!
Interactive list of the Algorithm finalists at the Competitive Programming Hall of Fame:
https://cphof.org/advanced/atcoder_wt/2026
Handles and ratings:
Damn, total domination by OpenAI's system in both, the heuristic and algorithmic tracks. I wonder what the world is going to look like a year from now.
Previously I thought chatGPT can defeat 99.99% competitive programmers, now I think it's 100% now, even tourist or jiangly cannot stand a chance.
then you need to compare with tourist clones working parallelly. :)
How was the AI setup to compete here? Was it solving all problems in parallel or sequentially? Is the feedback provided to the model same as for a participant(i.e. when it decides to submit, it is only told the verdict as WA/TLE or is there a human in the loop too?).
It was parallely solving all problems. A person was operating it, just doing submissions for GPT, and telling it back how many testcases passed/failed.
I would really like more details about the model used + costs to operate/processing power used. I think it's still pretty important to know these things when comparing a model to human participants.
https://x.com/i/status/2076789170706792661